Accuracy of segmentation of tooth structures using 3 different CBCT machines
Eman Shaheen, MSc, PhD,a,bWael Khalil, DDS, MSc,aMostafa Ezeldeen, DDS, MSc,a
Elke Van de Casteele, MSc, PhD,aYi Sun, MSc, PhD,a,bConstantinus Politis, MD, DDS, MHA, MM, PhD,a,band Reinhilde Jacobs, DDS, MSc, PhD, Dr hca,b,c
Objective. The aim of this study was to evaluate the volumetric accuracy and reliability of cone beam computed tomography (CBCT)-based tooth segmentation using 4 different CBCT exposure protocols.
Methods. Two dry, intact adult human mandibles of unknown gender were scanned using 4 different CBCT exposure protocols (3 CBCT systems). The available mandibular premolars (3 per mandible) were segmented, resulting in a total of 24 segmented teeth. To assess the accuracy of the segmented teeth, volumetric and morphologic differences between the real anatomic teeth and the reconstructed images were evaluated both physically and using a high-resolution micro-computed tomography system.
Results. Results revealed a high accuracy of CBCT reconstructed images when comparing volumetric measures of CBCT-based segmented premolars to physical measurements of corresponding physical teeth. Volumetric differences were below 2%.
Morphologic differences using the segmented model and the corresponding micro-computed tomography scans of the physical teeth indicated that when inaccuracies occurred, they were at the apical and coronal parts of the tooth.
Conclusion. Based on these results, CBCT can be used as a tool for segmentation and pretherapeutic planning procedures.
(Oral Surg Oral Med Oral Pathol Oral Radiol 2017;123:123-128)
During the last decade, 3-dimensional (3-D) information has been used more frequently to assist in dentomax- illofacial diagnostics and surgical planning. However, the use of computed tomography (CT) in daily dental practice remains contested due to its high cost and high radiation exposure.1Alternatively, cone beam computed tomography (CBCT) has demonstrated utility, with higher spatial resolution, smaller exposure, and lower cost compared to CT, but it still lacks contrast accuracy2 and evidence-based guidelines for a stan- dardized protocol.1,3
Three-dimensional virtual models obtained from CBCT images could be valuable tools for diagnosis and treatment planning. They could have a major impact on clinical practice, especially when combined with 3-D printing technology. However, their accuracy and effectiveness must be assessed before these models can be adopted.4CBCT accuracy has been studied in order to establish dimensional veriﬁcation, usually using osteologic landmarks on dry human skulls as reference points for measurement.5-8
In this context, the assessment should include all steps, starting from the accuracy of scanning to the segmentation procedure, the latter being a major step in creating accurate digital teeth to allow production of 3-D tooth models. However, to validate the accuracy of the resulting 3-D models, all steps must be taken into account. They include scanning, segmentation, and model fabrication, as previously validated and described by Shahbazian et al.9,10
The segmentation accuracy of CT has already been studied extensively.11In CT imaging, segmentation of objects or tissues is performed using thresholding based on prior knowledge of the density of the anatomic structure (Hounsﬁeld units). Unfortunately, gray values cannot be used directly in a quantitative way in CBCT imaging.2,12 In addition, low-contrast
aOMFS IMPATH Research Group, Department of Imaging and Pathology, Faculty of Medicine, Katholieke Universiteit Leuven, Leuven, Belgium.
bDepartment of Oral and Maxillofacial Surgery, University Hospitals Leuven, Leuven, Belgium.
cDepartment of Dental Medicine, Karolinska Institutet, Stockholm, Sweden.
Received for publication Apr 3, 2016; returned for revision Jul 28, 2016; accepted for publication Sep 15, 2016.
Ó 2016 Elsevier Inc. All rights reserved.
2212-4403/$ - see front matter
Statement of Clinical Relevance
The use of cone beam computed tomography (CBCT)-based tooth segmentation and replica fabri- cation may help to enhance therapeutic outcomes. A tooth replica may provide optimal bone ﬁt during tooth autotransplantation while reducing extra- alveolar time, thus preserving periodontal ligament and pulp vitality and reducing the risk of necrosis and resorption. Fabrication of the tooth replica is limited by the CBCT quality and the accuracy of the seg- mentation; therefore, studies exploring the inﬂuence of CBCT machines and protocols are highly relevant to this particular clinical application.
segmentation in CBCT is hampered by higher image noise compared with CT.13
Tooth segmentation is more challenging than bone structure segmentation for several reasons, such as the number of teeth per jaw, the proximity of adjacent tooth structures, the difference in density within a tooth (enamel, dentin, cementum, and pulp chamber), and tooth development.14 It is even more challenging to perform segmentation in CBCT images than in CT images. The studies that have considered the use of CBCT for tooth segmentation have been limited to only one system or protocol.9,10
The purpose of this study was to evaluate the volu- metric and morphologic accuracy and reliability of CBCT-based tooth segmentation, based on different CBCT units and varying exposure parameter protocols.
MATERIALS AND METHODS Study sample and collection
This study was carried out on 2 dry, intact adult human mandibles of unknown gender collected from the Department of Anatomy at KU Leuven. Ethical review board approval was obtained (ML9535/ML9248, ERB University Hospitals Leuven).
Image acquisition (cone beam computed tomography, micro-computed tomography) The mandibles were placed on a plastic tray with cop- perﬁlters of 0.5 mm in front of the X-ray beam source to simulate soft tissue and to reduce X-ray beam- hardening effects.15 Both dry dentate mandibles were scanned using 3 different CBCT machines (Table I):
Accuitomo 170 (Morita, Kyoto, Japan), Scanora 3-D (Soredex, Tuusula, Finland), and ProMax (Planmeca OY, Helsinki, Finland). For the Accuitomo 170, 2 different scanning protocols were used, half and full rotation. In total, 8 mandible scans were obtained. The 3 available mandibular premolars were selected from each of the 2 mandibles for further tooth segmentation, resulting in a total of 24 segmented teeth. All data sets were exported using the Digital Imaging and
Communications in Medicine (DICOM)ﬁle format. An example of an extracted premolar (tooth #21) from one of the dry mandibles is shown in Figure 1, with the corresponding segmented output.
All premolars were segmented from the DICOM images using the SimPlant Pro 3-D planning software (version 12.01, Dentsply Implants, Mölndal, Sweden) following a predeﬁned protocol.3The segmentation protocol has been previously validated9and explained in more detail.16
To assess segmentation accuracy, dimensional differ- ences between the real anatomic teeth (volume deter- mined using the Archimedes principle, as described by Khalil et al.16) and the segmented teeth were evaluated.
Table I. Speciﬁcations of the 4 cone beam computed tomography (CBCT) exposure protocols used for CBCT-based tooth segmentation
Accuitomo 170 180 Accuitomo 170 360 Scanora 3-D ProMax Max
Tube current (mA) 5 5 8 11
Gray scale (bit) 8 8 12 12
Potential (KV) 90 90 85 96
Scan time (s) 8 s/180 17.5 s/360 3.7 s/360 15 s/210
Reconstruction time (min) 5 5 1-2 3
Voxel size (mm) 0.16 0.16 0.2 0.15
Field of view (mm) 80 80 80 80 75 100 100 90
Detector type Flat panel Flat panel Flat panel Flat panel
Fig. 1. A, The physical tooth (#21) as it was extracted from the dry mandible. B, The segmented model of the same tooth.
In a second step, a micro-CT (
mCT) system was used for reference images because of its high resolution.
Volumetric comparisons were made between the segmented teeth from the different CBCT protocols and the corresponding segmentations from the
Each tooth was extracted from the dry mandible and scanned separately in the SkyScan 1172
(Bruker micro-CT, Kontich, Belgium). In a full rotation setting, with a voxel of size 17.8
mm, the source was set at 100 kV/100
mA, and an aluminum-copperﬁlter was used to reduce the beam-hardening effect. To increase the signal-to-noise ratio, the frame averaging was set at 6 frames per rotation step. Projection images were obtained over 180 with a rotation step of 0.7. The exposure time was 316 milliseconds, leading to a scan time of 12 minutes per scan. After acquisition, recon- struction was done using NRecon software (Bruker micro-CT). Reconstruction parameters were as follows:
Gaussian smoothing of 1, ring artifact correction of 7, and beam-hardening correction of 30%. After recon- struction, the teeth were segmented from the images to obtain 3-D surface-rendered models, and the corre- sponding volumes were calculated using CTAnalyser (Bruker micro-CT). Figure 2 shows an example of a premolar scanned with the
mCT system and the reconstructed volume in 3 orthogonal views (axial, coronal, and sagittal).
The 3-D models of the segmented teeth obtained from the CBCT and
mCT scans were registered using 3-matic software (version 9.0, Materialise NV, Leuven, Belgium). Morphologic errors and volumetric changes between the CBCT and
mCT models were measured via a point-based signed part comparison method, resulting in a color-coded map. This map expresses the distribution of the surface distance (Euclidean distance) between each point on the surface of the segmented tooth from CBCT and its corre- sponding point from
mCT. Distances greater than 0.25 mm are represented in red, differences of approximately zero are represented in green, and in- termediate distances are represented in orange. An example is shown inFigure 3.
All data were analyzed with the IBM Statistical Package for Social Sciences (SPSS, version 21.0, IBM Corpo- ration, Armonk, NY, USA). A comparison between the volume measurements of the physical teeth and the CBCT data sets was performed using repeated analysis of variance; the signiﬁcance level was set at P .05.
Pearson correlation was performed to examine the potential linear relationships. The degree of agreement between the volume measurements was compared using the Bland and Altman method.17
Fig. 2. Example of a premolar scanned with the microcomputed tomography (mCT) system and the reconstructed volume in 3 orthogonal views (axial, coronal, and sagittal).
The volume measurements of the
mCT revealed a strong positive correlation with those of the 4 CBCT protocols and with the physical measurements determined using the Archimedes principle (r > 0.90) when evaluated using the Pearson correlation (Table II).
The mean absolute difference in percentage between physical measurements determined using the Archi- medes principle and CBCT volume measurements was calculated using the Bland-Altman method and was found to be 1.9% with Accuitomo 170 180 rotation,
1.6% with Accuitomo 170 360 rotation, 2.1% with Promax Max, and 0.9% with Scanora 3-D (Table III).
The mean absolute difference between
CBCT volume measurements was found to be 3.6%
with Accuitomo 170 180 rotation, 3.2% with Accui- tomo 170 360 rotation, 3.8% with Promax Max, and 2.4% with Scanora 3-D (Table III).
Moreover, the mean absolute difference in percent- age between
mCT volume measurements and physical Fig. 3. Color-coded visualization in part comparison analysis showing localization of inaccuracies. Red represents large deviations (distances>0.25 mm), green represents small deviations (distances around 0 mm), and orange represents intermediate deviations between green and red.
Table II. Pearson coefﬁcients (R) of microcomputed tomography (
mCT) volume measurements, cone beam computed tomography (CBCT) volume measurements, and physical measurements determined using the Archimedes principle
R P value
Accuitomo 170 180rotation (virtual 3-D model)
Accuitomo 170 360rotation (virtual 3-D model)
Promax Max (virtual 3-D model) 0.92 .008y Scanora 3-D (virtual 3-D model) 0.93 .008y
Physical 0.91 .012*
*Signiﬁcance at P < .05.
ySigniﬁcance at P < .01.
Table III. Mean absolute difference in percentage of cone beam computed tomography (CBCT) volume measurements and physical measurements determined using the Archimedes principle and microcomputed tomography (
Mean absolute difference in percentage Physical volume and Accuitomo
Physical volume and Accuitomo 170 (360rotation)
Physical volume and Promax Max 2.1%
Physical volume and Scanora 3-D 0.9%
mCT and Accuitomo 170 180rotation 3.6%
mCT and Accuitomo 170 360rotation 3.2%
mCT and Promax Max 3.8%
mCT and Scanora 3-D 2.4%
mCT and Physical 1.6%
measurements determined using the Archimedes prin- ciple of the teeth was 1.6% (Table III).
The results of repeated analysis of variance revealed no statistically signiﬁcant differences between
and CBCT volume measurements (P ¼ .146).
Furthermore, no statistically signiﬁcant difference was found between
mCT and physical volume measure- ments (P¼ .489) (Table IV).
Morphologic analysis showed that 95% of the differ- ences between surfaces ranged from3.3 to þ1.5 mm, with a mean of 0.5 mm and a standard deviation of 0.8 mm. According to the color-coded map, higher deviations (red areas) were found in the apical region (18 models out of 24) and the coronal part (14 out of 24);
only a few models showed inaccuracies in the root part (Figure 3).
When looking at the future of 3-D printed medical tools and replicas as a support for clinical diagnosis, plan- ning, and treatment, it is of utmost importance to assess the accuracy of the virtual 3-D model obtained after segmentation. Since the CBCT data sets can be obtained from different scanners or scanning protocols, the present study assessed the accuracy and robustness of the obtained virtual 3-D models.
Volumetric measurements were made using the Archimedes principle, given its simplicity, consistency, and accuracy in measuring the volume of a given object with homogenous density.18 The mean volumetric differences for the Archimedes method and CBCT measurements varied in our study from 0.9% to 2.1%, which is consistent with the ﬁndings of studies conducted in similar settings.19-22 Compared with
mCT, the volume measurements and physical mea- surements were overestimated and varied from 1.9% to 3.6%. As previously calculated by Star et al.,23 the mean volume of the pulp is about 2% to 3% of the volume of the tooth. Therefore, the differences could be explained by the fact that the root canal was segmented in the
mCT images but not in the CBCT images, due to the difﬁculty of segmentation.
Furthermore, the lower resolution of CBCT (voxel
size range 0.15-0.2 mm) compared to
mCT (voxel size 0.0178 mm) may have contributed to this error. For correlational accuracy, it was shown that high correlation coefﬁcients were found between the segmented volumes and the original volumes in both applications (Archimedes and
mCT), which is also consistent with other studies.24
CBCT imaging quality is related to the machine settings, patient positioning, volume reconstruction, and DICOM export. These factors could affect the accuracy assessment. In the present study, data sets were collected from 3 different CBCT systems, with ﬁxed mandibles instead of patients who may move. This factor was not included in this study and thus is considered a limitation that should be tested in future work. Another possible factor that was not included in this study was the effect of artifacts, such as metal artifacts.25The presence of artifacts in the scan, whether in the neighborhood of the tooth of interest or not, would affect the gray values and thus the quality of the segmentation, which is based on thresholding.
The voxel sizes of the CBCT scan protocols used to scan teeth with the purpose of segmentation or diag- nostic evaluation fell approximately within the range covered in this study (0.15-0.2 mm). Even though no statistically signiﬁcant difference was found among all scanning protocols and the reference (whether
physical) for the covered range, larger voxel sizes were not considered based on theﬁndings of Maret et al.26In their study, underestimations were found for scans with a voxel size of 0.3 mm.
In the present study, results reveal that all tested CBCT protocols provided high accuracy for tooth segmenta- tion compared with anatomic tooth morphology.
Therefore, CBCT-segmented teeth can be recom- mended as a tool for diagnostic and pretherapeutic planning procedures.
1.Liang X, Jacobs R, Hassan B, et al. A comparative evaluation of cone beam computed tomography (CBCT) and multi-slice CT (MSCT), part I. On subjective image quality. Eur J Radiol.
2.Pauwels R, Jacobs R, Singer SR, Mupparapu M. CBCT-based bone quality assessment: are Hounsﬁeld units applicable?
Dentomaxillofac Radiol. 2015;44:20140238.
3.Loubele M, Maes F, Schutyser F, Marchal G, Jacobs R, Suetens P. Assessment of bone segmentation quality of cone- beam CT versus multislice spiral CT: a pilot study. Oral Surg Oral Med Oral Pathol Oral Radiol Endod. 2006;102:225-234.
4.Al-Rawi B, Hassan B, Vandenberge B, Jacobs R. Accuracy assessment of three-dimensional surface reconstructions of teeth from cone beam computed tomography scans. J Oral Rehabil.
Table IV. Comparison of microcomputed tomography (
mCT) volume measurements, cone beam computed tomography (CBCT) volume measurements, and physical measurements using analysis of variance (ANOVA) with repeated measures
Comparisons P value
mCT and CBCT (Accuitomo 170 180, Accuitomo 170 360, Promax, Scanora)
mCT and Archimedes’ volumetry .489
Note. Level of signiﬁcance set at P < .05.
5.Lagravre MO, Carey J, Toogood RW, Major PW. Three-dimensional accuracy of measurements made with software on cone-beam computed tomography images. Am J Orthod Dentofacial Orthop.
6.Loubele M, Van Assche N, Carpentier K, et al. Comparative localized linear accuracy of small-ﬁeld cone-beam CT and mul- tislice CT for alveolar bone measurements. Oral Surg Oral Med Oral Pathol Oral Radiol Endod. 2008;105:512-518.
7.Periago DR, Scarfe WC, Moshiri M, Scheetz JP, Silveira AM, Farman AG. Linear accuracy and reliability of cone beam CT derived 3-dimensional images constructed using an orthodontic volumetric rendering program. Angle Orthod. 2008;78:387-395.
8.Hassan B, van der Stelt P, Sanderink G. Accuracy of three-dimensional measurements obtained from cone beam computed tomography surface-rendered images for cephalometric analysis: inﬂuence of pa- tient scanning position. Eur J Orthod. 2009;31:129-134.
9.Shahbazian M, Jacobs R, Wyatt J, et al. Accuracy and surgical feasibility of a CBCT-based stereolithographic surgical guide aiding autotransplantation of teeth: in vitro validation. J Oral Rehab. 2010;37:854-859.
10.Shahbazian M, Jacobs R, Wyatt J, et al. Validation of the cone beam computed tomographyebased stereolithographic surgical guide aiding autotransplantation of teeth: clinical case-control study. Oral Surg Oral Med Oral Pathol Oral Radiol Endod.
11.Aamodt A, Kvistad KA, Andersen E, et al. Determination of Hounsﬁeld value for CT-based design of custom femoral stems.
J Bone Joint Surg Br. 1999;81:143-147.
12.Pauwels R, Nackaerts O, Bellaiche N, et al. Variability of dental cone beam CT grey values for density estimations. Br J Radiol.
13. Hiew LT, Ong SH, Foong KWC. Tooth Segmentation From Cone-Beam CT using graph cut. Proceedings of the Second APSIPA Annual Summit and Conference. December 14-17, 2010; Biopolis, Singapore.
14.Hui G, Oksam C. Individual tooth segmentation from CT images using level set method with shape and intensity prior. Pattern Recognition. 2010;43:2406-2417.
15.Suomalainen A, Vehmas T, Kortesniemi M, Robinson S, Peltola J. Accuracy of linear measurements using dental cone beam and conventional multislice computed tomography. Den- tomaxillofac Radiol. 2008;37:10-17.
16.Khalil W, EzEldeen M, Van De Casteele E, et al. Validation of cone beam computed tomographyebased tooth printing using different three-dimensional printing technologies. Oral Surg Oral Med Oral Pathol Oral Radiol Oral Endod. 2016;121:307-315.
17.Bland JM, Altman DG. Statistical methods for assessing agree- ment between two methods of clinical measurement. Lancet.
18.Hughes SW. Archimedes revisited: a faster, better and cheaper method of accurately measuring the volume of small objects.
Physics Education. 2005;40:468-474.
19.Agbaje JO, Jacobs R, Maes F, Michiels K, van Steenberghe D.
Volumetric analysis of extraction sockets using cone beam computed tomography: a pilot study on ex vivo jaw bone. J Clin Periodont. 2007;34:985-990.
20.EzEldeen M, Van Gorp G, Van Dessel J, Vandermeulen D, Jacobs R. CBCT-based bone quality assessment: are Hounsﬁeld units applicable? J Endod. 2015;41:317-324.
21.Liu Y, Olszewski R, Alexandroni ES, Enciso R, Xu T, Mah JK.
The validity of in vivo tooth volume determinations from cone- beam computed tomography. Angle Orthod. 2010;80:160-166.
22.Pinsky HM, Dyda S, Pinsky RW, Misch KA, Sarment DP. Ac- curacy of three-dimensional measurements using cone-beam CT.
Dentomaxillofac Radiol. 2006;35:410-416.
23.Kamburoglu K, Kolsuz E, Kurt H, Kiliç C, Özen T, Paksoy CS.
Accuracy of CBCT measurements of a human skull. J Digit Im- aging. 2011;24:787-793.
24.Star H, Thevissen P, Jacobs R, Fieuws S, Solheim T, Willems G.
Human dental age estimation by calculation of pulp-tooth volume ratios yielded on clinically acquired cone beam computed to- mography images of monoradicular teeth. J Forensic S. 2011;56:
25.Vasconcelos KF, Nicolielo LF, Nascimento MC, et al. Artefact expression associated with several cone-beam computed tomo- graphic machines when imaging root ﬁlled teeth. Int Endod J.
26.Maret D, Telmon N, Peters OA, et al. Effect of voxel size on the accuracy of 3 D reconstructions with cone beam CT. Dento- maxillofac Radiol. 2012;41:649-655.
Omfs Impath research group Department of Imaging and Pathology Faculty of Medicine
Katholieke Universiteit Leuven Kapucijnenvoer 33
3000 Leuven Belgium