Application of metal artifact reduction algorithm for CBCT diagnosis of temporary anchorage device–tooth root contact: inadequate to reduce false-positive rate

Abstract

Objectives:

It was recently found that when cone beam computed tomography (CBCT) was used to examine temporary anchorage device (TAD)–tooth root contact, it tends to yield high false-positive (FP) diagnoses. This study investigated whether application of a metal artifact reduction (MAR) algorithm or reducing CBCT scan voxel-size can remediate this problem.

Methods:

18 fresh pig cadaver mandibles underwent TAD placement bilaterally at first molar lingual furcation regions. CBCT scans were taken under varied MAR (absence, presence) and voxel-size (400 µm, 200 µm) settings. Then, TADs were removed and a micro-CT scan (27 µm voxel-size) of the TAD placement site was performed. Three raters, blinded of CBCT scan setting, independently diagnosed whether TADs were in contact with roots. The reliability and accuracy of CBCT diagnoses using micro-CT as the gold-standard were statistically examined.

Results:

Generally, CBCT diagnoses had intrarater (Cohen’s κ: 0.54–1) and interrater (Fleiss’ κ: 0.73–0.81) reliability, within the moderate to excellent range, which did not vary with MAR setting or scan voxel-size. For diagnostic accuracy, FP rate among all raters was mostly in the 15–25% range and did not change with MAR or scan voxel-size settings (McNemar tests, p > 0.05) while false-negative rate was relatively minimal and only occurred to one rater (9%).

Conclusions:

When using CBCT to diagnose possible TAD–root contact, applying a currently available Planmeca MAR algorithm or reducing CBCT scan voxel-size from 400 µm to 200 µm may not decrease FP rate. Further optimization of the MAR algorithm for this purpose may be needed.

Introduction

In orthodontics, TADs have become a widely used auxiliary modality for anchorage control. ¹ TADs need to remain stable to provide the intended clinical purpose. One factor that may compromises TAD stability is its close proximity to an adjacent tooth root. ² To accurately detect possible contact between the TAD and tooth root, some orthodontists and researchers have recently started using CBCT, especially when a root contact is suspected clinically. ³

Most TAD placement sites are near tooth roots, which are separated from surrounding alveolar bone by a thin layer of periodontal ligament (PDL). Combined, they constitute a structure with limited gray level contrast and spatial resolution. ⁴ Additionally, artifacts associated with metal objects on CBCT imaging often cause discrepancies between the reconstructed data and real objects. ⁵ As metal objects absorb a considerable amount of lower energy photons emitted from the X-ray source, they produce cupping and beam hardening artifacts appearing as distortion of the edges of metal structures and as dark streaks in the reconstructed images, respectively. ^5,6 These artifacts further challenge the diagnosis of TAD–root contact. Srinivasan et al. recently found that CBCT diagnosis tends to have a relatively high false-positive (FP) rate when comparing CBCT diagnosis of TAD–root contact with gold-standard micro-CT imaging, ⁷ to which they explained in detail how metal-related artifact and relatively low spatial resolution with 400 µm voxel-size scans may have contributed.

Conceptually, better control of these two possible factors may result in more accurate diagnosis. In order to reduce metal artifacts, CBCT scanner manufacturers started developing algorithms to reduce or even remove metal-associated artifacts. ⁸ One of such algorithms is the ARA algorithm developed by Planmeca (Planmeca Oy, Helsinki, Finland). Without being provided much detail about the mechanism of the algorithm, the users can choose to activate it before taking the CBCT scan or manually apply/adjust it during the image reconstruction process. Images presented by the company show that applying this algorithm does make metal in the reconstructed data appearing clearer and less ambiguous. The effectiveness of using such a MAR algorithm in improving diagnosis of TAD-root contact, however, has yet to be investigated. In terms of CBCT spatial resolution, one commonly used approach is to decrease the scan voxel-size. Whether decreasing CBCT voxel-size improves diagnostic accuracy of TAD-root contact, is also awaiting to be investigated. Therefore, the purpose of this study was to assess the effect of MAR application and reduced voxel-size scan on diagnostic accuracy for TAD-root proximity.

Methods and materials

Sample size calculation

Cadaver mandibles from 5- to 8-month-old domestic pigs (Sus scrofa), which closely resemble adolescent human dental and alveolar bone anatomy and function, ⁹ comprised the sample for this study. At this age, generally the domestic pig’s mandibular first permanent molar (M₁) has erupted while the adjacent fourth primary molar (p₄) remains intact and stable, which exfoliates at an older age (1.5–2 years). ¹⁰ Sample size determination was based on a finding from Srinivasan et al. that TAD–tooth root diagnostic accuracy averaged 70% for CBCT scans taken at 400 µm voxel-size on pig mandibles with a 69% prevalence, ⁷ and on another finding from Melo et al. that decreasing voxel-size from 300 to 200 µm increased the sensitivity and specificity of detecting longitudinal root fracture on teeth with posts by 27 and 3%, respectively. ¹¹ Based on the sensitivity, specificity and prevalence values reported in these two studies, ^7,11 we estimated that MAR application and/or reducing scanning voxel-size to 200 µm could improve the accuracy of diagnosing of TAD-root contact from 70 to 95%, representing a 25% improvement. To detect the difference between two proportion measurements, we calculated the required sample size using a Z-test in the G*Power program. ¹² The outcome was that a sample of 36 (TAD sites) was needed for each group to achieve an 80% power at an α = 0.05 level. As the same TAD sites were to be used under different settings (groups) of CBCT scan, the total sample size was thus determined as 36.

Sample characteristics

Eighteen freshly frozen, unembalmed pig heads were collected from a local pork processing company. After thawing, the mandibles were dissected from the head with adjacent soft tissue kept intact, resulting in a total of 36 hemimandibles with each hemimandible intended to contain one TAD site. Because the two TAD sites on a cadaver mandible were independent for this type of study, each TAD site instead of a mandible was considered a sample in this study, a design that has also been used in other CBCT studies investigating the effectiveness of MAR. ⁸ In addition, as no live animal experiments were performed and only post-mortem animal specimens from a food company were used, this study was not required for an IACUC review.

TAD placement and image acquisition

Each mandible first underwent a baseline CBCT scan at 400 µm voxel-size to visualize mesiodistal root angulation using a Planmeca Promax 3D machine (Planmeca Oy, Helsinki, Finland) for subsequent TAD placement. During scanning, the anterior mandible was placed facing forward, lower incisor midline centered and the occlusal plane parallel with the horizontal plane to simulate ideal clinical positioning. The midline and right/left retromolar region were scribed with a permanent marker to ensure the same positioning for subsequent CBCT scans. Due to the large size of pig mandibles, a 16 x 10 cm field of view (FOV) and a 180° partial rotation of the X-ray source and detector around the sample, was used for each scan.

Based on the initial CBCT, one 8.0 mm (length) by 1.4 mm (diameter) TAD of (Vector TAS-Ormco Corporation, Orange, CA) was placed on the lingual aspect below the furcation of the permanent first molar. This location was chosen based on a preliminary experiment on two specimens that showed the range of interroot distance at the mid-root level in the sagittal view to be 2.5–4 mm, similar to ideal TAD placement site in human patients. ¹³ The TAD was placed within keratinized gingiva above the mucogingival junction by one operator (VM), who was experienced with clinical TAD placement. The operator placed the self-drilling TAD manually with a driver matching the TAD head. Without pilot drilling, the operator applied steady manual torque to the TAD for the insertion and the process ended when the TAD collar was in touch with the gingiva.

Clinically, TADs may be placed at locations with or without brackets nearby. As it was previously found by Srinivasan et al. that metal brackets on adjacent tooth crowns did not make a difference to CBCT diagnosis of TAD–root contact, ⁷ this study did not further investigate that factor.

After TAD placement, four more CBCT scans were acquired: at 400 µm voxel-size with and without MAR application (90 kVp, 12.5 mA, 13.5 s) and at 200 µm voxel-size with and without MAR (90 kVp, 14 mA, 18 s). The MAR algorithm was a built-in module of the CBCT unit with activation instruction provided. Detail of the algorithm especially its mechanism in removing metal artifacts is not provided by the vendor. Upon completion of CBCT scan acquisition, the TAD was carefully unscrewed from each site. All soft tissue was removed from the pig mandible, then a 15 × 10 mm block containing the TAD placement site and adjacent tooth roots was sectioned with a diamond disc bur. The sectioned specimens subsequently underwent a micro-CT scan using SkyScan micro-CT scanner (SkyScan 1172 X-ray microtomograph, Antwerp, Belgium) at 27 µm voxel-size.

Image analysis

All CBCT and micro-CT scans were reconstructed into DICOM data and used for analysis in Dolphin 3D (Patterson Supply, Inc.) and ImageJ (National Institutes of Health, Bethesda, MD) software, respectively. All CBCT scans were randomly ordered and assigned ID codes to blind the scan settings to the raters. Three raters (VM, JL, KN), all with training in orthodontics, were calibrated by a board-certified radiologist (SK), and independently conducted CBCT analyses. Two raters (VM, JL) conducted the micro-CT analysis independently. The number of raters was mainly determined by following a design commonly adopted by many previous CBCT studies. ^14–16

Both CBCT and micro-CT scan analyses were completed in the sagittal and axial orthogonal planes with consistent orientation and slice thickness of 1 voxel. The x-axis was oriented parallel to the occlusal plane for axial analysis and the yaw was adjusted to orient the viewing plane perpendicularly to the TAD for sagittal analysis. After establishing proper scan orientation, qualitative diagnosis of TAD–tooth root contact was determined. More specifically, the entire length of the TAD was evaluated with no restrictions on zooming or altering the contrast to simulate a clinical scenario. TADs that overlapped an adjacent root were recorded as in-contact. TADs were recorded as without contact when alveolar bone was visible between a TAD and the adjacent tooth root. For specimens where TAD–tooth root contact was difficult to discern, the shortest distance between the TAD tip and adjacent tooth root was measured using a digital ruler tool of the program. Based on the average PDL thickness reported in humans ¹⁷ and pigs, ¹⁸ distances less than 300 µm were considered in-contact and those equal to or larger than 300 µm were considered out-of-contact.

In cases where the diagnosis differed between the sagittal and axial plane for a single specimen, the scan was reviewed and the rater came to an unambiguous diagnosis for each specimen. Finally, to evaluate intrarater reliability, all CBCT scans were randomly ordered again and assigned new ID codes and re-evaluated by the three raters independently after a one-month interval. All micro-CT scans were also re-evaluated independently by the two raters.

Interroot distance at the TAD level was also measured from 200 µm voxel-size CBCT images in the sagittal view of by one rater (ZS).

Statistical analysis

Intrarater reliability of micro-CT and CBCT diagnosis was assessed by Cohen’s κ tests. Interrater reliability of micro-CT and CBCT diagnosis was evaluated by Cohen’s and Fleiss’ κ tests, respectively. Interpretation of κ test results followed Kottner et al.’s guidelines. ¹⁹

Using micro-CT diagnoses as the gold-standard, the diagnostic parameters including sensitivity, specificity and accuracy for each rater were calculated. Then, the FP and false-negative (FN) rate of the diagnosis for each rater, were compared among MAR settings and voxel-size settings with McNemar tests. FP rate was defined as the proportion of cases falsely diagnosed as in-contact by CBCT among all cases diagnosed as out-of-contact by micro-CT. FN rate was defined as the proportion of cases falsely diagnosed as out-of-contact by CBCT among all cases diagnosed as in-contact by micro-CT. Finally, the positive-predictive value (PPV) and negative-predictive value (NPV) of each rater’s CBCT diagnoses were calculated. PPV was defined as the probability that an in-contact CBCT diagnosis was correct. NPV was defined as the probability that an out-of-contact CBCT diagnosis was correct.

Results

Representative CBCT and micro-CT images showing TADs in-contact and out-of-contact with adjacent roots are displayed in Figure 1. Overall, micro-CT images presented substantially higher tissue contrast and clarity of the TAD location. The dentition and eruption status shown in the CBCT images were also coherent with the reported age range of these pigs at the time of sacrifice. Even with the use of the initial CBCT as a reference by the same operator, TAD placement varied some in the vertical position relative to the furcation. Overall, interroot distance measured at the TAD level of all specimens measured from CBCT images in the sagittal view was 4.0 ± 0.6 mm.

Figure 1.

Representative CBCT and micro-CT images in coronal (left) and axial (right) views showing a TAD in-contact (a) and a TAD out-of-contact (b) with an adjacent root. Arrows indicate holes after TAD removal. Note that CBCT images have drastically lower clarity and contrast than micro-CT images. CBCT, cone beam CT; TAD, temporary anchorage device.

For diagnostic reliability, micro-CT diagnosis had perfect intrarater reliability (Cohen’s κ, 1.0), and nearly perfect interrater reliability (κ = 0.936). The only sample showing disagreement was further reviewed by a third evaluator (ZS) and a consensus was established. The same micro-CT diagnosis of all specimens were then used as the gold-standard. Overall, among a total of 36 TADs place, micro-CT analysis showed that 11 were in contact with roots, representing a 31% prevalence.

For CBCT diagnosis, intrarater reliability showed a large range (Cohen’s κ, 0.54–1. Figure 2A), while interrater reliability was generally good (Fleiss’ κ, 0.73–0.81, Figure 2B). Neither MAR nor voxel-size settings changed the reliability of CBCT diagnosis in any consistent pattern.

Figure 2.

Intra- (a) and interrater reliability (b) of CBCT diagnosis of TAD-root contact under varied conditions. MAR algorithm application. CBCT, cone beam CT; MAR, metal artifact reduction; TAD, temporary anchorage device.

Diagnostic sensitivity, specificity and accuracy of each rater during their first round of analysis are shown in Table 1. Overall, all raters had much higher sensitivity (range 91–100%) than specificity (range 72–92%) regardless of variations in voxel-size and MAR settings. The accuracy ranges for the three rater were 81–94%.

Table 1.

Diagnostic sensitivity, specificity and accuracy of 3 raters

CBCT scan settings	Rater 1			Rater 2			Rater 3
	Sensitivity	Specificity	Percentage accuracy	Sensitivity	Specificity	Percentage accuracy	Sensitivity	Specificity	Percentage accuracy
200 µm a	91%	84%	86%	100%	84%	89%	100%	80%	86%
200 µm + MAR b	91%	84%	86%	100%	92%	94%	100%	72%	81%
400 µm	91%	84%	86%	100%	76%	83%	100%	84%	89%
400 µm + MAR	91%	80%	83%	100%	92%	94%	100%	84%	89%

MAR, metal artifact reduction.

^aVoxel-size setting: 200 µm or 400 µm.

^bMAR algorithm application.

The FP rate of all raters is summarized in Figure 3, which was mostly over 15% except for Rater 2’s diagnoses under the MAR settings. McNemar tests among different MAR or voxel-size settings showed only one comparison reached statistical significance (p < 0.05). That was when Rater 2 analyzed 400 µm voxel-size CBCT scans with and without MAR applied (8% vs 24%, p < 0.05). In contrast, FN rate which is complementary to sensitivity (Table 1), was 9% for Rater 1 but 0% for the other two raters (details not shown).

Figure 3.

FP rate of all raters. MAR, with application of MAR algorithm. *, p < 0.05 based on McNemar test. FP, false-positive; MAR, metal artifact reduction.

Representative images showing the qualitative effects of MAR application on TADs in CBCT images are shown Figure 4. Overall, MAR application reduced metal artifacts to some extent including beam hardening and cupping as well as the size of the TADs.

Figure 4.

Representative images in coronal (left) and axial (right) views showing qualitative changes after application of MAR algorithm during CBCT image acquisition. CBCT, cone beam CT; MAR, metal artifact reduction.

The PPV and NPV of CBCT diagnosis for each rater’s first analysis are displayed in Table 2. Overall, NPV was nearly perfect while PPV was mostly below 80%. These measurements were generally similar to what was reported by Srinivasan et al., who only investigated one voxel-size setting (400 µm) without MAR. The application of MAR showed a tendency to improve the PPV to be above 80% in Rater 2 but not in the other two raters (Table 2).

Table 2.

PPV and NPVs

CBCT scan settings	PPVa				NPVb
	Rater 1	Rater 2	Rater 3	Srinivasan et al.	Rater 1	Rater 2	Rater 3	Srinivasan et al.
200 µmc	71%	73%	69%	NA	95%	100%	100%	NA
200 µm + MAR d	71%	85%	61%	NA	95%	100%	100%	NA
400 µm	71%	65%	73%	75%	95%	100%	100%	100%
400 µm + MAR	67%	85%	73%	NA	95%	100%	100%	NA

^aPPV, positive-predictive value.

^bNPV, negative-predictive value.

^cVoxel-size setting: 200 µm or 400 µm

^dMAR, metal artifact reduction algorithm application.

Discussion

Although CBCT imaging has been used by orthodontists to diagnose TAD-root contact proximity, especially when a direct contact is suspected, ¹ the reliability and accuracy of such diagnosis are not without problems, one of which is high FP rate (low specificity) likely associated with metal artifacts at the TAD–tooth root interface. ⁷ This study evaluated whether application of a MAR algorithm or smaller voxel-size scan could remediate this problem.

Similar to the Srinivasan et al study, ⁷ pig mandibular cadavers with intact soft tissue were used as the experimental model and micro-CT scans were adopted as the gold-standard for evaluation of CBCT diagnosis. Justification for using pig cadavers were detailed in their published article. ⁷ The rationale that micro-CT can be used as a gold-standard was also provided in their study ⁷ and in another article by Gulibire et al. ²⁰ The TAD placement location, however, was modified to better simulate clinical conditions. More specifically, in the Srinivasan study et al., a TAD was placed on the buccal alveolar bone between adjacent teeth, while in this study TADs were placed at the lingual furcation of the first permanent mandibular molar. Their location had a relatively narrow interroot distance. In comparison, the lingual furcation area where TADs were placed in this study had 4.0 ± 0.6 mm interroot distance. Although this site is rarely used for TADs in human patients, it has a clinically relevant interroot distance and alveolar bone thickness, thus representing an appropriate location for the purpose of this study. The difference of interroot distance at the TAD sites between this and the Srinivasan’s study impacted the prevalence of actual TAD–root contact. More specifically, the prevalence of positive TAD–root contact based on micro-CT diagnosis was higher (69%) when the interroot distance was narrower in the Srinivasan’s study than that in our study (31%), when the interroot distance was wider.

While the difference in prevalence can affect certain diagnostic parameters, the main diagnostic error identified in both studies are the same. ⁷ That is, diagnosis of TAD–root contact tends to have a relatively high FP rate (low specificity), while a low FN rate (high sensitivity) (Table 2). The relatively high FP rate generally did not change with MAR application or reduction of scan voxel with the only exception being Rater 2’s analysis of 400 µm voxel-size scans with or without MAR application (Figure 3, p < 0.05). During a second round of analysis, which was used to calculate intrarater reliability, this difference was no longer present. Because this statistical difference only appeared once and only occurred to one rater, it is very likely a Type I error. This explanation is further reflected by the PPV and NPV, parameters indicative of the probability of a positive and negative diagnosis, respectively, to match the gold-standard. More specifically, PPVs were below 80% in both studies despite that the FP rate is lower when TADs are placed at relatively wider locations (Table 2). Based on these PPVs, overall one can only expect a 70–80% probability that a positive diagnosis of TAD–root contact by CBCT is true. NPV, on the other hand, was nearly perfect in both studies (95%–100%), indicating that the certainty is strong when a negative TAD–root contact diagnosis is reached from CBCT imaging.

These data then refute the idea that FP rate in diagnosing TAD–root contact can be reduced by applying a currently available MAR algorithm, although, qualitatively, MAR-treated images do show smaller metal objects and less beam hardening artifacts (Figure 4). Why MAR was able to make the metal artifacts smaller but was not able to reduce FP rate of TAD–root diagnosis may be explained by a few factors?

First, the current MAR built-in algorithm is not optimized enough for the specific need of diagnosing TAD–root contact. On one hand, as we only investigated the Planmeca Promax 3D machine and its built-in MAR in this study, our finding may only indicate that this particular CBCT machine and MAR algorithm need further optimization for clinical diagnosis of TAD–root contact. On the other hand, our finding is not unprecedented. Kamburoglu et al., ²¹ who evaluated periimplant and periodontal defects with the use of MAR and did not find it effective in improving the diagnosis. Similarly, a study by Dalili Kajan et al. evaluated the effect of MAR on vertical root fracture diagnosis in endodontically treated teeth and found that diagnostic accuracy was not significantly improved with MAR application, etc.. ²² Yet another study by Bechara et al. actually found that MAR application even decreased the diagnostic accuracy of detecting root fractures in endodontically treated teeth.. ²³ In the latter two studies, another CBCT machine, Master 3D^® (Vatech, Hwaseong, Republic of Korea), and its built-in MAR were used. As the MAR algorithm is proprietary property of the companies, it is unknown how similar the MAR algorithms developed by different companies are, but together, these findings indicate that more MAR algorithm optimization may be needed for diagnosing fine structures in the dental field.

Second, the spatial resolution of the bone–PDL–root interface in CBCT may be too low for human eyes to discern the exact location of the TAD tip even without the interference of metal artifacts, when it is very close to the interface. With the thickness of the PDL reported to be 150–350 µm in humans ¹⁷ and on average 360 µm in pigs, ¹⁸ the anatomic separation between the alveolar bone and the root cementum, which are of similar gray level, is probably no more than 300–400 µm. This sets up the clinical goal when it comes to diagnosing whether a TAD encroaches the PDL. Meanwhile, the spatial resolution offered by CBCT imaging is theoretically the size of a voxel but realistically double the size of a voxel. ²⁴ This latter point is corroborated by the finding that the use of smaller voxel-size (200 µm) does not offer remediation for the relatively high FP rate, either. It is worth noting that experienced raters may be able to guess the scan voxel-size based on the resolution of the image, which would make them unblinded of the CBCT setting, which might increase a rater anticipation bias for finding a difference between the two scan settings. As our data did not show a difference between the two voxel-sizes, this potential bias is probably minimal in this study. In contrast to our finding, several previous studies did show benefits of smaller voxel-sizes (200 µm vs 400 µm) in detecting simulated temporomandibular joint erosion and mesiobuccal canal identification, but not in detecting simulated root resorption. ^14,25,26 Putting their reports and our finding together, we reason that reducing voxel-size from 400 to 200 µm may help the diagnosis of structures or lesions with relatively large anatomical separation form adjacent tissues, but is ineffective in improving diagnosis of the narrow bone–PDL–root interface. Although this ineffectiveness may be due to CBCT imaging being inherently inferior to medical CT and micro-CT or due to inadequate reduction of the voxel-size, reducing voxle-size from 400 to 200 µm offers no tangible diagnostic/clinical benefits for TAD placement while increasing radiation exposure patients. Thus, it is not justified according to the ALADAIP (acceptable being indication-oriented and patient-specific) principle. ²⁷

One more possible explanation is that the presence of even the slightest metal artifacts in the immediate field overrides any potential improvement from reduced voxel-size. Consistent with this notion, Codari et al. evaluated the effect of 13 different CBCT machines with various FOVs and voxel-size settings on the expression of metal artifacts using high density metal phantoms (amalgam, titanium, and copper–aluminum alloy). Their results showed no significant difference in the expression of metal artifacts with voxel-sizes ranging from from 80 to 400 µm. ²⁸ If this is confirmed to be the case for TADs placed near the bone–PDL–root interface, complete metal artifact resolution in CBCT images is required to eliminate FP diagnosis.

In addition to diagnostic accuracy, our data did not find that MAR or decreasing voxel-size improved diagnostic reliability in any consistent way (Figure 2). For interrater reliability, it is worth confirming that having three raters instead of two is a legitimate approach to improve the statistical power of the κ test without inflating the actual measurement of reliability. ²⁹ In numerous previous CBCT studies, 2–3 raters have been adopted which provides empirical evidence for our design. Although some other studies included four or more raters, ^20,22 it remains unclear whether a statistical gold-standard for deciding the number of raters exist. Nevertheless, based on our current study design, the differences of intrarater reliability among raters may be related to their previous experience with interpreting CBCT images. Of the three raters in this study, one (JL) had more research experience with CBCT or CT imaging than the other two. On the other hand, given that the interrater reliability was fairly similar among varied scan settings, it is more likely that the image quality changes caused by MAR application or voxel-size reduction is not significant enough to impact the diagnosis of TAD-root contact. The findings suggests that other unknown factors besides the metal artifacts and low spatial resolution may be contributing to the inaccuracy in diagnosing TAD-root contact with CBCT.

In essence, from a clinical point of view, the findings from this study continue to raise caution for using CBCT to ascertain TAD–root contact especially when clinically a contact is suspected. Neither MAR nor reduction of scan voxel-size is able to decrease the FP rate. While one can still use MAR since it does not involve additional harm, the use of small voxel-size scans needs to be avoided as it typically comes with a higher radiation exposure. ³⁰

As with other studies, this study has its limitations. First, we used an FOV that is larger than that typically used for TADs on orthodontic patients. This was chosen to accommodate the size of the pig mandible. As large FOVs tend to obscure finer structures, it may be a potential confounding factor. Second, clinically CBCT may only be used when the clinician was suspecting that a TAD in close proximity to a root, especially when the patient complains about pain on the adjacent teeth. Using cadaver tissue, this feedback was certainly missing from this study, thus TADs far from roots and TADs in close proximity to roots were all lumped together for the analysis. Thus the FP rate (15–25%) found in this study may be underestimated compared to a real clinical situation.

Conclusions

Even with the application of a MAR algorithm or reducing scan voxel-size from 400 to 200 µm, diagnosis of TAD–root contact from CBCT images may still have some FP risk. Future efforts are needed to further optimize the MAR algorithm for the purpose of improving diagnosis of TAD–root contact using CBCT imaging.