Automatic Pulp and Teeth Three-Dimensional Modeling of Single and Multi-Rooted Teeth Based on Cone-Beam Computed Tomography Imaging: A Promising Approach With Clinical and Therapeutic Outcomes

Abstract

Background Cone-beam computed tomography (CBCT) imaging offers high-quality three-dimensional (3D) acquisition with great spatial resolution, given by the use of isometric voxels, when compared with conventional computed tomography (CT). The current literature supports a median reduction of 76% (up to 85% reduction) of patients' radiation exposure when imaged by CBCT versus CT. Clinical applications of CBCT imaging can benefit both medical and dental professions. Because these images are digital, the use of algorithms can facilitate the diagnosis of pathologies and the management of patients. There is pertinence to developing rapid and efficient segmentation of teeth from facial volumes acquired with CBCT. Methodology In this paper, a segmentation algorithm using heuristics based on pulp and teeth anatomy as a pre-personalized model is proposed for both single and multi-rooted teeth. Results A quantitative analysis was performed by comparing the results of the algorithm to a gold standard obtained from manual segmentation using the Dice index, average surface distance (ASD), and Mahalanobis distance (MHD) metrics. Qualitative analysis was also performed between the algorithm and the gold standard of 78 teeth. The Dice index average for all pulp segmentation (n = 78) was 83.82% (SD = 6.54%). ASD for all pulp segmentation (n = 78) was 0.21 mm (SD = 0.34 mm). Pulp segmentation compared with MHD averages was 0.19 mm (SD = 0.21 mm). The results of teeth segmentation metrics were similar to pulp segmentation metrics. For the total teeth (n = 78) included in this study, the Dice index average was 92% (SD = 13.10%), ASD was low at 0.19 mm (SD = 0.15 mm), and MHD was 0.11 mm (SD = 0.09 mm). Despite good quantitative results, the qualitative analysis yielded fair results due to large categories. When compared with existing automatic segmentation methods, our approach enables an effective segmentation for both pulp and teeth. Conclusions Our proposed algorithm for pulp and teeth segmentation yields results that are comparable to those obtained by the state-of-the-art methods in both quantitative and qualitative analysis, thus offering interesting perspectives in many clinical fields of dentistry.

Keywords: cbct; dental segmentation; image processing and analysis; pulpal segmentation; segmentation algorithm.

Figure 1. Flowchart showing a summary of the methodology.

Figure 2. Coronal view of a molar (1.6) showing the intermediate steps involved in the pulp segmentation algorithm.

(A) Original image. (B) The difference between A and hole-filled A. (C) A binary marker created from B. (D) A smoothed with a Gaussian filter. (E) D after black top hat. (F) Hole-filled E image (mask). (G) Final pulp segmentation (after masking reconstruction).

Figure 3. Area evolution model and its application to cut an apical segmentation overflow.

(A) Model with a gradient and margin of two standard deviations. B (i): Cut of the apical overflow and (ii) analyzed region. (C) Segmentation before the cut. (D) Segmentation after the cut.

Figure 4. For (A) pulp segmentation and (B) tooth segmentation, examples of categories for (i) good, (ii) crown overflow, (iii) apical overflow, (iv) missing canal, and (v) missing canal extremity.

Figure 5. Three-dimensional rendering of segmented (A) pulp and tooth 2.4. (B) Overlapping of expert’s segmentation (dark blue) and segmentation algorithm result (light blue).