doi:10.1111/iej.12477
Performance of an artefact reduction algorithm in the diagnosis of in vitro vertical root fracture in four different root filling conditions on CBCT images
G. L. de Rezende Barbosa1, S. L. Sousa Melo2, P. N. B. Alencar1, M. C. C. Nascimento1 & S. M. Almeida1 1
Division of Radiology, Department of Oral Diagnosis, Piracicaba Dental School, University of Campinas, S~ ao Paulo, Brazil; and Department of Oral Pathology, Radiology and Medicine, University of Iowa College of Dentistry, Iowa City, IA, USA
2
Abstract de Rezende Barbosa GL, Sousa Melo SL, Alencar PNB, Nascimento MCC, Almeida SM. Performance of an artefact reduction algorithm in the diagnosis of in vitro vertical root fracture in four different root filling conditions on CBCT images. International Endodontic Journal, 49, 500–508, 2016.
Aim To evaluate the influence of an artefact reduction algorithm (ARA) and several root filling materials on the detection of root fractures on cone-beam computed tomography (CBCT) images. Methodology Forty-four teeth were divided into control and fractured groups and scanned on a Picasso Trio CBCT device under four conditions: unrestored, filled with gutta-percha cones, cast-gold or fibreglass posts; either with or without applying the ARA. Three calibrated examiners assessed the images. ROC analysis, ANOVA and pairwise Tukey LSD test were performed.
Introduction The increasing use of CBCT in dentistry has raised concerns. Radiation protection regulation agencies have emphasized that patients must be only exposed to a dose as low as diagnostically achievable (Shaw & Cro€ uail 2013, Farman 2014). In other words, if a more advanced image technique exposes patients to
Correspondence: Saulo L. Sousa Melo, 801 Newton rd. S367 Dental Science Bld. Iowa City, IA 52246, USA (Tel.: +1 (319) 335 9656; e-mail: [email protected]).
500
International Endodontic Journal, 49, 500–508, 2016
Results No significant difference between the groups with and without the ARA was observed. There was no significant interaction between the algorithm and root condition. On the other hand, there was a significant difference in the mean values of sensitivity (Sn) and accuracy (Ac) amongst the different root filling groups (P ≤ 0.001). Conclusions The application of the ARA did not influence the diagnosis of root fractures, and its effects did not depend on root conditions. In relation to the filling materials, gold posts reduced the overall CBCT diagnostic ability, regardless of the use of the ARA. Keywords: algorithms, cone-beam CT, diagnostic tests, root canal filling materials, root fracture. Received 28 March 2015; accepted 26 May 2015
higher doses of radiation, its application can only be justified if the diagnosis has not been satisfactorily achieved by conventional lower dose radiographs. A definitive diagnosis of root fracture (RF) is one of the most challenging tasks in dentistry. Usually, it is a dilemma: a misdiagnosis can lead to a subclinical periodontal disease that has the potential to exacerbate over time or an overdiagnosis can result in unnecessary tooth extraction (Kajan & Taromsari 2012). It is well established that two-dimensional radiographic techniques only allow the visualization of the fracture line if the X-ray beam has been oriented along that line. Otherwise, overlapping of adjacent
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
de Rezende Barbosa et al. Artefact reduction algorithm
structures will obscure the RF, leading to misdiagnoses (Nair et al. 2001, Kamburoglu et al. 2009, Tsesis et al. 2010, Junqueira et al. 2013, Brady et al. 2014). Thus, cone-beam computed tomography (CBCT) might be a better alternative as it associated with greater accuracy compared with periapical radiographs (Bernardes et al. 2009, Hassan et al. 2009, Kamburoglu et al. 2009, Wenzel et al. 2009). The most common causes of RF are trauma and restoration of the tooth. The former usually leads to horizontally oriented fractures with displacement of fragments. The latter is associated with hairline vertical/longitudinal fractures (Fuss et al. 2001, Cohen et al. 2003). As these teeth often undergo restorative procedures with highly dense materials, such as gutta-percha, metal posts and crowns, the visualization of fracture lines in CBCT images may be jeopardized due to image artefacts (e.g. streaking and/ or beam hardening artefacts) (Hassan et al. 2009, 2010, Melo et al. 2010, Khedmat et al. 2012, Junqueira et al. 2013, Patel et al. 2013, da Silveira et al. 2013, Brady et al. 2014). Depending on the type of the root filling, the artefacts will vary in prominence (Melo et al. 2010, Brito-Junior et al. 2014). Several authors have studied various means to improve the CBCT image quality for the visibility of the fracture line by testing different devices, datareconstruction parameters and/or software (Junqueira et al. 2013, Melo et al. 2013, Avsever et al. 2014, Brito-Junior et al. 2014). Moreover, some CBCT units allow the application of modified algorithms (e.g. artefact reduction algorithm – ARA) during image reconstruction in the hope of enhancing the contrast-to-noise ratio (CNR) of the acquired
(a)
(b)
images (Choi et al. 2011, Wu et al. 2011, Bechara et al. 2012a,b). It is hypothesized that the application of the ARA would improve the accuracy of RF detection due to the reduction of artefacts present in the image. This study aimed to evaluate the influence of an ARA on the accuracy of detection of simulated RF in root filled teeth. In addition, the influence of guttapercha cones, cast-gold and fibreglass posts was evaluated. The null hypothesis was that the ARA did not improve the overall diagnostic performance in any of the studied simulations.
Materials and methods This study protocol was independently reviewed and approved by the Institutional Review Board of the University of Campinas and is in compliance with the Helsinki Declaration. Fifty single-rooted human teeth were selected and inspected by transillumination for the absence of root fracture, and by periapical radiographs to observe root canal anatomy (Fig. 1a). Teeth with caries extending to the root, restorations, pulp calcifications, root resorption or root fracture were excluded. The crown was removed at the cemento-enamel junction to facilitate root fracture induction, eliminate the bias of enamel fractures and identify the sample. Access openings were performed and the root canals were instrumented using the ProTaper rotary system (Dentsply Maillefer, Ballaigues, Switzerland) up to size F5 (Fig. 1b). Finally, each tooth was randomly coded and divided into control nonfractured and artificially fractured groups (NF and AF, respectively).
(c)
(d)
Figure 1 Steps of tooth preparation prior to CBCT scanning. (a) periapical radiograph to observe root canal anatomy; (b) standardization of the shape of the root canal; (c) fracture induction; (d) inspection via transillumination to confirm the presence of root fracture (adapted from Melo et al. 2013).
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
International Endodontic Journal, 49, 500–508, 2016
501
Artefact reduction algorithm de Rezende Barbosa et al.
The AF group teeth were coated with a layer of wax and fixed in a mini-table lathe. Longitudinal RF was induced by a conical wedge with a bevelled tip driven apically into each tooth, and controlled pressure being applied by gentle tapping until a complete fracture was achieved (Fig. 1c), as described previously (Melo et al. 2010). The root fragments were replaced in their original position to simulate the immediate post-trauma situation. The artificially fractured teeth were reinspected by transillumination to confirm the presence of a hairline, uniform, longitudinal RF (Fig. 1d). Six teeth were used to understand the force that was needed to break the root into only two fragments and were excluded from the final sample. The final sample comprised 22 nonfractured and 22 artificially fractured teeth. Each of the 44 teeth of the final sample was scanned under four different root canal conditions: unrestored (Control), filled with gutta-percha cone (Gutta-percha), restored with castgold post (Gold post) and restored with fibreglass post (Fibreglass post). The fillings were passively well fitted and noncemented in the root canal of the teeth with care not to displace the fragments. Cementation was avoided to prevent the cement from flowing into the fracture line and obscuring it on the image. A ProTaper F5 gutta-percha cone, a gold alloy post or a fibreglass post (Angelus, Londrina, Brazil) were used. Prior to scanning, each sample was coated with a layer of wax (approximately 1 mm thick) and randomly placed in the premolar, canine or incisor sockets of a dry human mandible, 6 or 7 teeth at a time. The mandible was coated with a 5-mm-thick piece of wax to simulate soft tissues in the image. For the acquisition of CBCT images, the sample was scanned in a Picasso Trio tomography device (E-WOO, Giheung-gu, Republic of Korea), according to the exposure protocol recommended by the manufacturer for high-resolution 0.2-mm voxel scans (80 kVp, 4 mA), either applying or not an ARA available on the acquisition software (EasyDent4, E-WOO, Giheung-gu, Republic of Korea). The exposure protocol was exactly the same when the ARA was applied; however, the inherent reconstruction time demanded by the software was twice as long. The ARA identifies the inconsistencies in the sinogram due to discrepancies in the linear attenuation coefficients and, by linear interpolation, replaces the sonogram corresponding to metallic parts with only interpolated values from the boundaries (Choi et al. 2011). Figure 2 illustrates two examples with all possible combinations in which the sample was scanned,
502
International Endodontic Journal, 49, 500–508, 2016
considering the variables included in the study: the presence of root fracture, root canal status and application of the ARA. Note that the same tooth was scanned 8 times, depending on the set-up. Figure 3 presents one tooth of the AF group visualized in all 8 possible combinations of root canal status and algorithms. The 352 resultant CBCT DICOM volumes (44 teeth 9 4 root canal status 9 2 image modalities) were imported into OnDemand3D software (Cybermed, Seoul, South Korea) for evaluation. The images were reconstructed preserving their acquisition features, without compression, coded and shown in a random order under dim-light conditions. Three blinded, previously calibrated oral and maxillofacial radiologists with at least 5 years of experience in CBCT diagnosis were the observers who performed the evaluation of the images on a 24-inch LCD monitor (MDRC-2124; Barco, Duluth, GA, USA) with a matrix resolution of 1920 9 1200. The calibration consisted of the identification of RF in tomograms that did not belong to the study, after providing instructions about CBCT image interpretation and usage of the software. The observers were told to evaluate the presence or absence of root fracture in all teeth included in the volume. However, they were unaware of the 50/50 probability of a root fracture being present. The evaluation was made by performing a dynamic reading of all orthogonal slices (axial, coronal and sagittal). For that, the images were orthogonally resliced in contiguous slice increments, starting with the realignment of the tooth long axis parallel to the sagittal plane. Adjustment of brightness, contrast and zoom were allowed, but task-specific filters were not permitted. The observers rated the presence or absence of RF on a 5-point scale as follows: (i) fracture definitely not present, (ii) fracture probably not present, (iii) uncertain whether fracture is present or not, (iv) fracture probably present and (v) fracture definitely present. The same observation was repeated after a 15-day interval for observer agreement evaluation. Cohen’s kappa was used to calculate intra- and interobserver agreement and interpreted as poor (0– 0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80) or almost perfect (0.81+). The diagnostic values of sensitivity (Sn), specificity (Sp) and accuracy (Ac) were performed by pooling observer responses for every image algorithm and root canal status. The accuracy was also assessed by receiver operating characteristic (ROC) analysis; the area
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
de Rezende Barbosa et al. Artefact reduction algorithm
Figure 2 Schematic diagram of all possible combinations of the presence/absence of root fracture, root canal status and application/nonapplication of the artefact reduction algorithm (ARA) in the sample.
under ROC curve (Az) and the optimal threshold point (cut-off value) from each ROC curve were calculated. The Sn, Sp, Ac and Az values were compared by two-way repeated-measures ANOVA. Pairwise comparisons of values of two different image algorithms under four root filling conditions were performed using Tukey LSD test. Data analyses were performed using SigmaStat for Windows (Systat Software, Erkrath, Germany). The level of significance was set at P < 0.05.
Results Intra-observer coefficients indicated almost perfect agreement (range 0.84–0.90), whilst interobserver values indicated substantial agreement (range 0.66–0.70). Table 1 summarizes the results for overall Sn, Sp and Ac for the diagnosis of vertical RF using or not ARA per experimental root filling group. Based on the observed means, there was no significant difference between the algorithms used (Sn, P = 0.807; Sp, P = 0.797; Ac, P = 1.000). The effect of image algorithm did not depend on what root filling condition was present; that is, there was not a significant interaction
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
between the algorithm and root canal status. On the other hand, there was a significant difference in the mean values of Sn and Ac amongst the different root filling groups (P ≤ 0.001). The Tukey test was used to isolate which groups differed from the others. Table 2 and Fig. 4 show the mean areas under ROC curves (Az) and the optimal threshold point for the observers in each image algorithm and root filling condition. In general, there was no significant difference amongst algorithms (P = 0.429). The effect of the algorithm did not depend on what material was present; that is, there was no significant interaction between the algorithm and root canal status. However, there was a significant difference amongst materials (P = 0.005). The Tukey test was used to isolate which groups differed from the others.
Discussion CBCT has been demonstrated to be superior to other radiographic modalities in the detection of RF (Bornstein et al. 2009, Hassan et al. 2009, Iikubo et al. 2009, Kamburoglu et al. 2009, Wenzel et al. 2009, Ozer 2010). The diagnostic capability of CBCT was
International Endodontic Journal, 49, 500–508, 2016
503
Artefact reduction algorithm de Rezende Barbosa et al.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
Figure 3 Final reconstructed axial images of the same tooth from the artificially fractured group scanned with each of the eight possible combinations with the standard algorithm in the top row and the artefact reduction algorithm (ARA) in the bottom row: a/e, unrestored (Control) canal; b/f, canal filled with gutta-percha cone; c/g, canal restored with cast-gold post; d/h, canal restored with fibreglass post. Arrows indicate the limits of detectable fracture lines.
Table 1 Sensitivity, specificity and accuracy of each image algorithm per experimental root filling group
Group Control Gutta-percha Gold post Fibreglass post Overall
Original X
0.89 0.69Y,Z 0.54Z 0.83X,Y 0.74
Sensitivity
Specificity
Accuracy
Artefact Reduction Algorithm
Artefact Reduction Algorithm
Artefact Reduction Algorithm
X
0.83 0.60Y 0.52Y 0.92X 0.72
P-value 0.565 0.448 0.844 0.448 0.807
Original 0.87 0.86 0.75 0.84 0.83
0.84 0.84 0.84 0.83 0.84
P-value 0.731 0.863 0.197 0.863 0.797
Original X,Y
0.88 0.79Y,Z 0.67Z 0.84X,Y 0.80
X,Y
0.84 0.76X,Y 0.73Y 0.86X 0.80
P-value 0.387 0.383 0.238 0.591 1.000
Means with different letters are statistically different (P < 0.05).
also proved to be superior for the detection of suspected RF in patients, revealing the possibility of translating the positive results into clinical practice (Edlund et al. 2011). However, it is also well known that the presence of artefacts decreases the image quality by reducing the contrast, obscuring structures and impairing the detection of areas of interest, thereby making the diagnosis difficult and time-
504
International Endodontic Journal, 49, 500–508, 2016
consuming (Barrett & Keat 2004, Berg et al. 2006). The detection of RF is one of the diagnostic tasks most affected by the presence of imaging artefacts (Costa et al. 2011, Khedmat et al. 2012). For this reason, several methods have been studied and proposed in the attempt to reduce artefacts in the image and improve its diagnostic capability. One of these is the application of the ARA.
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
de Rezende Barbosa et al. Artefact reduction algorithm
Table 2 Mean areas under ROC curves (Az), and threshold point, of each image algorithm per experimental root filling group Original Group
Az
Control Gutta-percha Gold post Fibreglass post
Artefact Reduction Algorithm
Threshold point
X
Az
(0.10,0.83) (0.15,0.66) (0.30,0.60) (0.13,0.78)
0.90 0.74X,Y 0.66Y 0.93X
Threshold point
P-value
(0.14,0.79) (0.25,0.63) (0.25,0.54) (0.13,0.83)
0.741 0.868 0.567 0.564
X
0.93 0.78X,Y 0.69Y 0.88X
Means with different letters are statistically different (P < 0.05). Standard error of mean SEM = 0.0562.
1
1
(a)
(b)
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0 0
0.2 Control
0.4 Gua-percha
0.6 Gold post
0.8
1
Fiberglass post
0
0.2 Control
0.4 Gua-percha
0.6 Gold post
0.8
1
Fiberglass post
Figure 4 Receiver operating characteristic curves for all observers in each experimental root filling group based on image algorithm: (a) standard, (b) artefact reduction algorithm (ARA). X-axis represents the false-positive rate (1 – specificity). Y-axis represents the true-positive rate (sensitivity). The line of no discrimination – diagonal line from the left bottom to the top right corners – represents a completely random guess.
According to the present results, the application of an ARA did not improve the detection of RF, regardless of the filling material. Likewise, no significant decrease in diagnosis was observed, contrary to a previous study that reported a reduction in the accuracy of RF detection in root filled teeth when the ARA from two different CBCT units was applied (Bechara et al. 2013). The hypothesis of improvement of RF detection was formulated due to the improvement of image quality by ARA (Bechara et al. 2012a,b). The algorithm appears to reduce the effects of beam hardening and scattering caused by metallic structures, increase the contrast-to-noise ratio and decrease the grey value variation. This is achieved by the interpolation-based sonogram corrections where the projection data of the metal objects is replaced by linear interpolation of surrounding data in the sinogram. The major drawback is the incomplete artefact correction or
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
even production of secondary artefacts (Wu et al. 2011). A previous study reported that the algorithm negatively changed the image, in that a part of the periphery of the gutta-percha was missing compared to the image without ARA and that it did not properly reduce the artefact or lead to a more accurate diagnosis (Bechara et al. 2013). In relation to the different root filling materials evaluated, a significantly lower accuracy of RF detection in canals filled with gutta-percha cones, despite the application of the ARA, was observed. Previous studies also identified a lower sensitivity and accuracy in the presence of gutta-percha (Khedmat et al. 2012, Bechara et al. 2013). On the other hand, several studies did not report a significant decrease in diagnostic ability (Melo et al. 2013). The accuracy was also lower in the presence of metallic posts, similar to previous findings (Melo et al. 2010, Costa et al. 2011). The gold post group had the lowest values in the
International Endodontic Journal, 49, 500–508, 2016
505
Artefact reduction algorithm de Rezende Barbosa et al.
diagnostic tests, having a significant lower sensitivity in comparison with the control and fibreglass post groups, regardless of the use of the ARA. No significant differences were observed in the specificity for the detection of RF in teeth with gold post group, which differs from previous studies that reported a more expressive influence of metallic posts in the specificity than in the sensitivity (da Silveira et al. 2013). The influence of intraradicular posts is understandable inasmuch as they are high-density materials that usually cause artefacts on CBCT images, interfering with diagnostic quality and decreasing the ability of depiction (Sanders et al. 2007). The difference found between posts (in the present study, gold and fibreglass) can also be explained by the different densities of the materials. Gold posts consist of high-density materials that cause more beam hardening and streak artefacts than the less dense fibreglass posts (Boas & Fleischmann 2011). If high-density materials impair the detection of areas of interest in the image and obscure anatomical structure (Barrett & Keat 2004), materials with lower interference in the image should improve the tomographic follow-up of these teeth. Inasmuch as the use of posts in certain root filled teeth is unquestionable and fundamental, future research should be conducted to improve and develop filling materials that combine the desired prosthetic effects with the lowest image interference possible. A limitation of CBCT images in detecting fine RF is related to its inherent spatial resolution and the partial volume averaging artefact. This cone-beamrelated artefact occurs when the size of the voxel (i.e. voxel resolution) is greater than the spatial or contrast resolution of the object to be imaged (Scarfe & Farman 2008). The protocol recommended by the manufacture of the CBCT device used in the present study advocates that the voxel resolution at 0.2 mm would provide a spatial resolution of 2.5 linepair mm 1. These resolutions may not be low enough to visualize all fractures that were simulated in the study. A previous study that took the thickness of the fracture line in account concluded that CBCT presented lower Sn and Az values for the detection of incomplete fractures, especially those of
Performance of an artefact reduction algorithm in the diagnosis of in vitro vertical root fracture in four different root filling conditions on CBCT images
G. L. de Rezende Barbosa1, S. L. Sousa Melo2, P. N. B. Alencar1, M. C. C. Nascimento1 & S. M. Almeida1 1
Division of Radiology, Department of Oral Diagnosis, Piracicaba Dental School, University of Campinas, S~ ao Paulo, Brazil; and Department of Oral Pathology, Radiology and Medicine, University of Iowa College of Dentistry, Iowa City, IA, USA
2
Abstract de Rezende Barbosa GL, Sousa Melo SL, Alencar PNB, Nascimento MCC, Almeida SM. Performance of an artefact reduction algorithm in the diagnosis of in vitro vertical root fracture in four different root filling conditions on CBCT images. International Endodontic Journal, 49, 500–508, 2016.
Aim To evaluate the influence of an artefact reduction algorithm (ARA) and several root filling materials on the detection of root fractures on cone-beam computed tomography (CBCT) images. Methodology Forty-four teeth were divided into control and fractured groups and scanned on a Picasso Trio CBCT device under four conditions: unrestored, filled with gutta-percha cones, cast-gold or fibreglass posts; either with or without applying the ARA. Three calibrated examiners assessed the images. ROC analysis, ANOVA and pairwise Tukey LSD test were performed.
Introduction The increasing use of CBCT in dentistry has raised concerns. Radiation protection regulation agencies have emphasized that patients must be only exposed to a dose as low as diagnostically achievable (Shaw & Cro€ uail 2013, Farman 2014). In other words, if a more advanced image technique exposes patients to
Correspondence: Saulo L. Sousa Melo, 801 Newton rd. S367 Dental Science Bld. Iowa City, IA 52246, USA (Tel.: +1 (319) 335 9656; e-mail: [email protected]).
500
International Endodontic Journal, 49, 500–508, 2016
Results No significant difference between the groups with and without the ARA was observed. There was no significant interaction between the algorithm and root condition. On the other hand, there was a significant difference in the mean values of sensitivity (Sn) and accuracy (Ac) amongst the different root filling groups (P ≤ 0.001). Conclusions The application of the ARA did not influence the diagnosis of root fractures, and its effects did not depend on root conditions. In relation to the filling materials, gold posts reduced the overall CBCT diagnostic ability, regardless of the use of the ARA. Keywords: algorithms, cone-beam CT, diagnostic tests, root canal filling materials, root fracture. Received 28 March 2015; accepted 26 May 2015
higher doses of radiation, its application can only be justified if the diagnosis has not been satisfactorily achieved by conventional lower dose radiographs. A definitive diagnosis of root fracture (RF) is one of the most challenging tasks in dentistry. Usually, it is a dilemma: a misdiagnosis can lead to a subclinical periodontal disease that has the potential to exacerbate over time or an overdiagnosis can result in unnecessary tooth extraction (Kajan & Taromsari 2012). It is well established that two-dimensional radiographic techniques only allow the visualization of the fracture line if the X-ray beam has been oriented along that line. Otherwise, overlapping of adjacent
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
de Rezende Barbosa et al. Artefact reduction algorithm
structures will obscure the RF, leading to misdiagnoses (Nair et al. 2001, Kamburoglu et al. 2009, Tsesis et al. 2010, Junqueira et al. 2013, Brady et al. 2014). Thus, cone-beam computed tomography (CBCT) might be a better alternative as it associated with greater accuracy compared with periapical radiographs (Bernardes et al. 2009, Hassan et al. 2009, Kamburoglu et al. 2009, Wenzel et al. 2009). The most common causes of RF are trauma and restoration of the tooth. The former usually leads to horizontally oriented fractures with displacement of fragments. The latter is associated with hairline vertical/longitudinal fractures (Fuss et al. 2001, Cohen et al. 2003). As these teeth often undergo restorative procedures with highly dense materials, such as gutta-percha, metal posts and crowns, the visualization of fracture lines in CBCT images may be jeopardized due to image artefacts (e.g. streaking and/ or beam hardening artefacts) (Hassan et al. 2009, 2010, Melo et al. 2010, Khedmat et al. 2012, Junqueira et al. 2013, Patel et al. 2013, da Silveira et al. 2013, Brady et al. 2014). Depending on the type of the root filling, the artefacts will vary in prominence (Melo et al. 2010, Brito-Junior et al. 2014). Several authors have studied various means to improve the CBCT image quality for the visibility of the fracture line by testing different devices, datareconstruction parameters and/or software (Junqueira et al. 2013, Melo et al. 2013, Avsever et al. 2014, Brito-Junior et al. 2014). Moreover, some CBCT units allow the application of modified algorithms (e.g. artefact reduction algorithm – ARA) during image reconstruction in the hope of enhancing the contrast-to-noise ratio (CNR) of the acquired
(a)
(b)
images (Choi et al. 2011, Wu et al. 2011, Bechara et al. 2012a,b). It is hypothesized that the application of the ARA would improve the accuracy of RF detection due to the reduction of artefacts present in the image. This study aimed to evaluate the influence of an ARA on the accuracy of detection of simulated RF in root filled teeth. In addition, the influence of guttapercha cones, cast-gold and fibreglass posts was evaluated. The null hypothesis was that the ARA did not improve the overall diagnostic performance in any of the studied simulations.
Materials and methods This study protocol was independently reviewed and approved by the Institutional Review Board of the University of Campinas and is in compliance with the Helsinki Declaration. Fifty single-rooted human teeth were selected and inspected by transillumination for the absence of root fracture, and by periapical radiographs to observe root canal anatomy (Fig. 1a). Teeth with caries extending to the root, restorations, pulp calcifications, root resorption or root fracture were excluded. The crown was removed at the cemento-enamel junction to facilitate root fracture induction, eliminate the bias of enamel fractures and identify the sample. Access openings were performed and the root canals were instrumented using the ProTaper rotary system (Dentsply Maillefer, Ballaigues, Switzerland) up to size F5 (Fig. 1b). Finally, each tooth was randomly coded and divided into control nonfractured and artificially fractured groups (NF and AF, respectively).
(c)
(d)
Figure 1 Steps of tooth preparation prior to CBCT scanning. (a) periapical radiograph to observe root canal anatomy; (b) standardization of the shape of the root canal; (c) fracture induction; (d) inspection via transillumination to confirm the presence of root fracture (adapted from Melo et al. 2013).
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
International Endodontic Journal, 49, 500–508, 2016
501
Artefact reduction algorithm de Rezende Barbosa et al.
The AF group teeth were coated with a layer of wax and fixed in a mini-table lathe. Longitudinal RF was induced by a conical wedge with a bevelled tip driven apically into each tooth, and controlled pressure being applied by gentle tapping until a complete fracture was achieved (Fig. 1c), as described previously (Melo et al. 2010). The root fragments were replaced in their original position to simulate the immediate post-trauma situation. The artificially fractured teeth were reinspected by transillumination to confirm the presence of a hairline, uniform, longitudinal RF (Fig. 1d). Six teeth were used to understand the force that was needed to break the root into only two fragments and were excluded from the final sample. The final sample comprised 22 nonfractured and 22 artificially fractured teeth. Each of the 44 teeth of the final sample was scanned under four different root canal conditions: unrestored (Control), filled with gutta-percha cone (Gutta-percha), restored with castgold post (Gold post) and restored with fibreglass post (Fibreglass post). The fillings were passively well fitted and noncemented in the root canal of the teeth with care not to displace the fragments. Cementation was avoided to prevent the cement from flowing into the fracture line and obscuring it on the image. A ProTaper F5 gutta-percha cone, a gold alloy post or a fibreglass post (Angelus, Londrina, Brazil) were used. Prior to scanning, each sample was coated with a layer of wax (approximately 1 mm thick) and randomly placed in the premolar, canine or incisor sockets of a dry human mandible, 6 or 7 teeth at a time. The mandible was coated with a 5-mm-thick piece of wax to simulate soft tissues in the image. For the acquisition of CBCT images, the sample was scanned in a Picasso Trio tomography device (E-WOO, Giheung-gu, Republic of Korea), according to the exposure protocol recommended by the manufacturer for high-resolution 0.2-mm voxel scans (80 kVp, 4 mA), either applying or not an ARA available on the acquisition software (EasyDent4, E-WOO, Giheung-gu, Republic of Korea). The exposure protocol was exactly the same when the ARA was applied; however, the inherent reconstruction time demanded by the software was twice as long. The ARA identifies the inconsistencies in the sinogram due to discrepancies in the linear attenuation coefficients and, by linear interpolation, replaces the sonogram corresponding to metallic parts with only interpolated values from the boundaries (Choi et al. 2011). Figure 2 illustrates two examples with all possible combinations in which the sample was scanned,
502
International Endodontic Journal, 49, 500–508, 2016
considering the variables included in the study: the presence of root fracture, root canal status and application of the ARA. Note that the same tooth was scanned 8 times, depending on the set-up. Figure 3 presents one tooth of the AF group visualized in all 8 possible combinations of root canal status and algorithms. The 352 resultant CBCT DICOM volumes (44 teeth 9 4 root canal status 9 2 image modalities) were imported into OnDemand3D software (Cybermed, Seoul, South Korea) for evaluation. The images were reconstructed preserving their acquisition features, without compression, coded and shown in a random order under dim-light conditions. Three blinded, previously calibrated oral and maxillofacial radiologists with at least 5 years of experience in CBCT diagnosis were the observers who performed the evaluation of the images on a 24-inch LCD monitor (MDRC-2124; Barco, Duluth, GA, USA) with a matrix resolution of 1920 9 1200. The calibration consisted of the identification of RF in tomograms that did not belong to the study, after providing instructions about CBCT image interpretation and usage of the software. The observers were told to evaluate the presence or absence of root fracture in all teeth included in the volume. However, they were unaware of the 50/50 probability of a root fracture being present. The evaluation was made by performing a dynamic reading of all orthogonal slices (axial, coronal and sagittal). For that, the images were orthogonally resliced in contiguous slice increments, starting with the realignment of the tooth long axis parallel to the sagittal plane. Adjustment of brightness, contrast and zoom were allowed, but task-specific filters were not permitted. The observers rated the presence or absence of RF on a 5-point scale as follows: (i) fracture definitely not present, (ii) fracture probably not present, (iii) uncertain whether fracture is present or not, (iv) fracture probably present and (v) fracture definitely present. The same observation was repeated after a 15-day interval for observer agreement evaluation. Cohen’s kappa was used to calculate intra- and interobserver agreement and interpreted as poor (0– 0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80) or almost perfect (0.81+). The diagnostic values of sensitivity (Sn), specificity (Sp) and accuracy (Ac) were performed by pooling observer responses for every image algorithm and root canal status. The accuracy was also assessed by receiver operating characteristic (ROC) analysis; the area
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
de Rezende Barbosa et al. Artefact reduction algorithm
Figure 2 Schematic diagram of all possible combinations of the presence/absence of root fracture, root canal status and application/nonapplication of the artefact reduction algorithm (ARA) in the sample.
under ROC curve (Az) and the optimal threshold point (cut-off value) from each ROC curve were calculated. The Sn, Sp, Ac and Az values were compared by two-way repeated-measures ANOVA. Pairwise comparisons of values of two different image algorithms under four root filling conditions were performed using Tukey LSD test. Data analyses were performed using SigmaStat for Windows (Systat Software, Erkrath, Germany). The level of significance was set at P < 0.05.
Results Intra-observer coefficients indicated almost perfect agreement (range 0.84–0.90), whilst interobserver values indicated substantial agreement (range 0.66–0.70). Table 1 summarizes the results for overall Sn, Sp and Ac for the diagnosis of vertical RF using or not ARA per experimental root filling group. Based on the observed means, there was no significant difference between the algorithms used (Sn, P = 0.807; Sp, P = 0.797; Ac, P = 1.000). The effect of image algorithm did not depend on what root filling condition was present; that is, there was not a significant interaction
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
between the algorithm and root canal status. On the other hand, there was a significant difference in the mean values of Sn and Ac amongst the different root filling groups (P ≤ 0.001). The Tukey test was used to isolate which groups differed from the others. Table 2 and Fig. 4 show the mean areas under ROC curves (Az) and the optimal threshold point for the observers in each image algorithm and root filling condition. In general, there was no significant difference amongst algorithms (P = 0.429). The effect of the algorithm did not depend on what material was present; that is, there was no significant interaction between the algorithm and root canal status. However, there was a significant difference amongst materials (P = 0.005). The Tukey test was used to isolate which groups differed from the others.
Discussion CBCT has been demonstrated to be superior to other radiographic modalities in the detection of RF (Bornstein et al. 2009, Hassan et al. 2009, Iikubo et al. 2009, Kamburoglu et al. 2009, Wenzel et al. 2009, Ozer 2010). The diagnostic capability of CBCT was
International Endodontic Journal, 49, 500–508, 2016
503
Artefact reduction algorithm de Rezende Barbosa et al.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
Figure 3 Final reconstructed axial images of the same tooth from the artificially fractured group scanned with each of the eight possible combinations with the standard algorithm in the top row and the artefact reduction algorithm (ARA) in the bottom row: a/e, unrestored (Control) canal; b/f, canal filled with gutta-percha cone; c/g, canal restored with cast-gold post; d/h, canal restored with fibreglass post. Arrows indicate the limits of detectable fracture lines.
Table 1 Sensitivity, specificity and accuracy of each image algorithm per experimental root filling group
Group Control Gutta-percha Gold post Fibreglass post Overall
Original X
0.89 0.69Y,Z 0.54Z 0.83X,Y 0.74
Sensitivity
Specificity
Accuracy
Artefact Reduction Algorithm
Artefact Reduction Algorithm
Artefact Reduction Algorithm
X
0.83 0.60Y 0.52Y 0.92X 0.72
P-value 0.565 0.448 0.844 0.448 0.807
Original 0.87 0.86 0.75 0.84 0.83
0.84 0.84 0.84 0.83 0.84
P-value 0.731 0.863 0.197 0.863 0.797
Original X,Y
0.88 0.79Y,Z 0.67Z 0.84X,Y 0.80
X,Y
0.84 0.76X,Y 0.73Y 0.86X 0.80
P-value 0.387 0.383 0.238 0.591 1.000
Means with different letters are statistically different (P < 0.05).
also proved to be superior for the detection of suspected RF in patients, revealing the possibility of translating the positive results into clinical practice (Edlund et al. 2011). However, it is also well known that the presence of artefacts decreases the image quality by reducing the contrast, obscuring structures and impairing the detection of areas of interest, thereby making the diagnosis difficult and time-
504
International Endodontic Journal, 49, 500–508, 2016
consuming (Barrett & Keat 2004, Berg et al. 2006). The detection of RF is one of the diagnostic tasks most affected by the presence of imaging artefacts (Costa et al. 2011, Khedmat et al. 2012). For this reason, several methods have been studied and proposed in the attempt to reduce artefacts in the image and improve its diagnostic capability. One of these is the application of the ARA.
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
de Rezende Barbosa et al. Artefact reduction algorithm
Table 2 Mean areas under ROC curves (Az), and threshold point, of each image algorithm per experimental root filling group Original Group
Az
Control Gutta-percha Gold post Fibreglass post
Artefact Reduction Algorithm
Threshold point
X
Az
(0.10,0.83) (0.15,0.66) (0.30,0.60) (0.13,0.78)
0.90 0.74X,Y 0.66Y 0.93X
Threshold point
P-value
(0.14,0.79) (0.25,0.63) (0.25,0.54) (0.13,0.83)
0.741 0.868 0.567 0.564
X
0.93 0.78X,Y 0.69Y 0.88X
Means with different letters are statistically different (P < 0.05). Standard error of mean SEM = 0.0562.
1
1
(a)
(b)
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0 0
0.2 Control
0.4 Gua-percha
0.6 Gold post
0.8
1
Fiberglass post
0
0.2 Control
0.4 Gua-percha
0.6 Gold post
0.8
1
Fiberglass post
Figure 4 Receiver operating characteristic curves for all observers in each experimental root filling group based on image algorithm: (a) standard, (b) artefact reduction algorithm (ARA). X-axis represents the false-positive rate (1 – specificity). Y-axis represents the true-positive rate (sensitivity). The line of no discrimination – diagonal line from the left bottom to the top right corners – represents a completely random guess.
According to the present results, the application of an ARA did not improve the detection of RF, regardless of the filling material. Likewise, no significant decrease in diagnosis was observed, contrary to a previous study that reported a reduction in the accuracy of RF detection in root filled teeth when the ARA from two different CBCT units was applied (Bechara et al. 2013). The hypothesis of improvement of RF detection was formulated due to the improvement of image quality by ARA (Bechara et al. 2012a,b). The algorithm appears to reduce the effects of beam hardening and scattering caused by metallic structures, increase the contrast-to-noise ratio and decrease the grey value variation. This is achieved by the interpolation-based sonogram corrections where the projection data of the metal objects is replaced by linear interpolation of surrounding data in the sinogram. The major drawback is the incomplete artefact correction or
© 2015 International Endodontic Journal. Published by John Wiley & Sons Ltd
even production of secondary artefacts (Wu et al. 2011). A previous study reported that the algorithm negatively changed the image, in that a part of the periphery of the gutta-percha was missing compared to the image without ARA and that it did not properly reduce the artefact or lead to a more accurate diagnosis (Bechara et al. 2013). In relation to the different root filling materials evaluated, a significantly lower accuracy of RF detection in canals filled with gutta-percha cones, despite the application of the ARA, was observed. Previous studies also identified a lower sensitivity and accuracy in the presence of gutta-percha (Khedmat et al. 2012, Bechara et al. 2013). On the other hand, several studies did not report a significant decrease in diagnostic ability (Melo et al. 2013). The accuracy was also lower in the presence of metallic posts, similar to previous findings (Melo et al. 2010, Costa et al. 2011). The gold post group had the lowest values in the
International Endodontic Journal, 49, 500–508, 2016
505
Artefact reduction algorithm de Rezende Barbosa et al.
diagnostic tests, having a significant lower sensitivity in comparison with the control and fibreglass post groups, regardless of the use of the ARA. No significant differences were observed in the specificity for the detection of RF in teeth with gold post group, which differs from previous studies that reported a more expressive influence of metallic posts in the specificity than in the sensitivity (da Silveira et al. 2013). The influence of intraradicular posts is understandable inasmuch as they are high-density materials that usually cause artefacts on CBCT images, interfering with diagnostic quality and decreasing the ability of depiction (Sanders et al. 2007). The difference found between posts (in the present study, gold and fibreglass) can also be explained by the different densities of the materials. Gold posts consist of high-density materials that cause more beam hardening and streak artefacts than the less dense fibreglass posts (Boas & Fleischmann 2011). If high-density materials impair the detection of areas of interest in the image and obscure anatomical structure (Barrett & Keat 2004), materials with lower interference in the image should improve the tomographic follow-up of these teeth. Inasmuch as the use of posts in certain root filled teeth is unquestionable and fundamental, future research should be conducted to improve and develop filling materials that combine the desired prosthetic effects with the lowest image interference possible. A limitation of CBCT images in detecting fine RF is related to its inherent spatial resolution and the partial volume averaging artefact. This cone-beamrelated artefact occurs when the size of the voxel (i.e. voxel resolution) is greater than the spatial or contrast resolution of the object to be imaged (Scarfe & Farman 2008). The protocol recommended by the manufacture of the CBCT device used in the present study advocates that the voxel resolution at 0.2 mm would provide a spatial resolution of 2.5 linepair mm 1. These resolutions may not be low enough to visualize all fractures that were simulated in the study. A previous study that took the thickness of the fracture line in account concluded that CBCT presented lower Sn and Az values for the detection of incomplete fractures, especially those of
