INTRODUCTION

Cone-beam computed tomographic (CBCT) was first introduced to dentistry in Europe since 1998 and approved for the clinical use in the USA in 2001 (^{Kapila & Nervina, 2015}). CBCT imaging has been proved to be a valuable tool in dentistry and been utilized for diagnosis, treatment planning and research (^{Merrett et al., 2009}; ^{Shahbazian et al., 2013}; ^{Park et al., 2015}). CBCT scanner has been used in the fields of oral surgery, endodontics, orthodontics and prosthodontics (^{Porciúncula et al., 2014}; ^{Liebregts et al., 2015}; ^{John et al., 2016}). CBCT images could be displayed in sagittal, axial and coronal planes for viewing teeth, bone, temporomandibular joint and other tissues.

Orthodontic treatment involves complex tooth movements in three-dimensional directions. Tooth root morphology, size and location are important for orthodontists. Using CBCT images, the anatomical information of mandibular canal, root and crown can be reconstructed (^{Kovisto et al., 2011}).

In addition, physicians would be able to perform more accurate treatments with the help of CBCT images, and keep the periodontal tissues healthy, or maintain crowns and roots in their new positions after the treatment. Without being given consideration, the dental root may move out of the bone cortex during or after the treatment (^{Cheng et al., 2015}). Therefore tooth segmentation was important for Orthodontic.

The tooth segmentation is a challenging work, as the lack of margins in the crown in CBCT images. Due to the noise of the image or the distribution density of the root and alveolar images are similar, it is difficult to segment the root (^{Gan et al., 2015}). ^{Heo & Chae (2004}) performed a tooth-by-tooth segmentation using the region growing algorithm, where the reference slice was first selected from CT slices and tooth segmentation was then carried out interactively based on shape characteristics of each tooth. The size, location and intensity of corresponding tooth were very similar, so it needed to define the accurate contour of each tooth form CT slices. ^{Hiew et al. (2010}) proposed a graph cuts segmentation approach to obtain the 3D tooth model from CBCT images. The results showed the roots of the tooth had some non-dental tissues. Due to low image contrast, higher image noise and missing image boundaries, tooth segmentation in CBCT is difficult.

The present study explores a new semi-automatic method to segment the teeth from the CBCT images. For evaluation, the CBCT-segmented tooth model is compared with the light-scanned tooth surface model.

MATERIAL AND METHOD

2.1 Tools: (1) Software: ORS Visual 2.0 (ORS Company, Canada), Geomagic Studio 2012 & Geomagic Qualify 2012 (3D Systems Company, America), (2) Hardware: structured white light scanner (IScan D104i, Imetric 3D SA, Switzerland).

2.2 data processing

2.2.1 Data acquisition. One male patient was selected as the experimental subject. The maxillary CBCT scans were performed on the CBCT machine (DCT Pro; Vatech & EWOO Group, South Korea). Images were obtained using the following protocol: field of view, 200 x 190 mm²; 90 kVp, 144 mA; scan time, 24 s; voxel size, 0.4 mm. Scanning conditions were constant with 360 rotation. All data were saved in the DICOM 3.0 format.

A structured white light scanner was used to obtain the 3D data for the upper, lower dental cast models and the relations to the jaw. The accuracy of the scanner was at 20mm. The scanned data was saved as ‘Model_Upper.stl’ and ‘Model_Lower.stl’ (Fig.1).

Fig.1 3D data of the upper and lower dental casts and their relation to the jaw. 

2.2.2 Segmentation. ORS Visual 2.0 software could provide a clinical workflow with selectable post-processing settings and real-time volumetric visualization, it also offers segmentation tools to visually isolate anatomy or features. We used this software to segment different teeth to different ROIs (Region of interests) and then analyze it. Then the data was saved as ‘Teeth_upper.stl’ and ‘Teeth_Lower.stl’. ‘Teeth_Upper.stl’ and ‘Teeth_Lower.stl’ were merged into one data called ‘Teeth.stl’. Software interfa- ce of the ORS software was showed in Figure 2.

Fig. 2 Software interface of the ORS. 

The proposed segmentation methodology consists of five stages: (a) loading data to the ORS software, (b) remove the data of the skull except the data upper and lower jaws, (c) based on gray value, segment teeth and save it as different ROIs, (d) combine different ROIs into an integrated mesh file, (e) smooth the mesh data. All the segmented teeth were shown in Figure 3.

Fig. 3 Segmented teeth. 

2.2.3 Registration. The teeth data (which meant ‘Teeth.stl’) and the model data (which meant ‘Model_Upper.stl’ and ‘Model_Lower.stl’) were both saved in STL format. The registration was linear transformation, which include rotation, translation. With the software of Geomagic Studio 2012, using the ‘Manual Registration’ settings, fixed the teeth data, and aligned the models to teeth data, so the stl data of teeth and the models were in a same coordinate system. Results were shown in Figure 4.

2.2.4 Deviation analysis. A deviation analysis was done to compare the teeth data and the model data. Geomagic Qualify 2012 software was used to analyze differences between the teeth data and the model data. The teeth data was set as ‘reference model’ and the model data was set as ‘test model’. The function of ‘Deviation analysis’ was used to analysis the distance from the Test to any point on the Reference.

Fig. 4 Cast models and the teeth data after registration in Geomagic Studio 2012. 

RESULTS

All teeth crown and root information can be accurately obtained using semi-automatic method to segment the tooth from the three-dimensional volume data which acquired from cone beam computed tomography. The method is simple to use and can be applied in orthodontic. The entire segmentation process took less than 30 minutes.

After registering the model data to teeth data, the deviation between them was then analyzed in Geomagic Studio. The actual values of the distance were -0.27 and 0.22 mm; the measured Root Mean Square (RMS) value was 0.39 mm, less than 0.4 mm. The deviation analysis result was shown in Figure 5.

Fig. 5 The deviation analysis result. 

DISCUSSION

By this method, all teeth crown data can be obtained and showed in reverse engineering software and can be applied in orthodontic. The entire process took less than 30 minutes.

CBCT imaging could provide detailed images of the bone and be performed to evaluate diseases of the jaw, dentition, bony structures of the face, nasal cavity, nerve canals and sinuses. Although it couldn’t provide the full diagnostic information available with conventional CT, particularly in evaluation of soft tissue structures such as muscles, lymph nodes, glands and nerves. But CBCT has the advantage of lower radiation exposure compared to conventional CT (^{Scarfe & Farman, 2008}).

In orthodontics, understanding the location of the root is very important for the orthodontic doctor. Diagnosis model based on crown cannot absolutely ensure the good alignment of roots without root exposure. It is necessary to construct the integrated model including root and crown for the diagnosis during tooth arrangement process. Tooth movement depended on the stress distribution in the periodontal ligament, whether it occurs through bone, or with bone were both important for the treatment (^{Melsen, 1999}). Diagnosis model based only on crown which could not absolutely ensure the good alignment of roots without root exposure (^{Bovali et al., 2014}; ^{Park et al., 2016}). It is necessary to construct the integrated model including root for diagnosis during tooth arrangement process. ^{El-Timamy et al. (2016}) used CBCT imaging and computer-aided manufacturing to produce stereolithographic trays for indirect-direct bonding, all the bracket positioning were based on the root axes, but the segmentation result with Mimics software still should be considered, and the segmentation accuracy of the teeth should be evaluated.

Threshold segmentation of CBCT was performed to generate 3D digital models (^{Machado, 2015}). In order to validate the accuracy of the segmented results, the upper and lower dental cast models were scanned by a white light scanner. The accuracy of the scanner was 20 mm. Than registration was done between the model data and teeth data, different registration method will affect the results, manual registration was used to manually align the two scans, 3 point-pairs on each of the data sets were selected. The deviation between them was analyzed in Geomagic Qualify. The actual values of the distance were -0.27 and 0.22 mm; the measured Root Mean Square (RMS) value is 0.39 mm, less than 0.4 mm.

There are several drawbacks to the software that need to be improved. First, the whole segmentation process depends on commercial software and requires human interaction, which reduces efficiency. In the future, we hope special software can be developed to enable an automated process. The CBCT can automatically be reconstructed into three-dimensional images for different structure. Second, the method couldn’t automatically determine the outer edge of the teeth margin. Root contour may be blurred because of the image noise and image intensity similarities between root and surrounding alveolar bone, especially in the posterior position, so the root extraction method should be improved and the operator should have a basic knowledge of dental anatomy. Third, it is impossible to intelligently identify teeth names and locations. Thus, further research is still underway to solve the above problems.

CONCLUSION

The semi-automatic method of segmenting the teeth from CBCT was realized. By the method, all teeth crown and root can be obtained. However, the whole design result needs to be precisely modified using the tools available in the two software packages to obtain a good result. Additionally, the entire design process was completed through human-computer interaction. Therefore, further studies are necessary to carry out and find an automated process software to make tooth segmentation design both simpler and faster.