relu

Link to paper

A novel deep learning system for multi-class tooth segmentation and classification on cone beam computed tomography. A validation study

‍

Eman Shaheen ^ab, André Leite ^bc, Khalid Ayidh Alqahtani ^bd, Andreas Smolders ^e, Adriaan Van Gerven ^e, Holger Willems ^e, Reinhilde Jacobs ^b ^f

^aDepartment of Oral and Maxillofacial Surgery, University Hospitals Leuven, Kapucijnenvoer 33, BE-3000 Leuven, Belgium

‍^bOMFS IMPATH Research Group, Department of Imaging and Pathology, Faculty of Medicine, KU Leuven, Kapucijnenvoer 33, BE-3000 Leuven, Belgium

‍^cDepartment of Dentistry, Faculty of Health Sciences, University of Brasília, Brasília, Brazil

‍^dDepartment of Oral and Maxillofacial Surgery and Diagnostic Sciences, College of Dentistry, Sattam Bin Abdulaziz University, Al-Kharj, Saudi Arabia^eRelu BV, Kapeldreef 60, BE-3001, Leuven, Belgium

‍^fDepartment of Dental Medicine, Karolinska Institutet, Box 4064, 141 04 Huddinge, Sweden

‍

Abstract

Objectives

Automatic tooth segmentation and classification from cone beam computed tomography (CBCT) have become an integral component of the digital dental workflows. Therefore, the aim of this study was to develop and validate a deep learning approach for an automatic tooth segmentation and classification from CBCT images.

Methods

A dataset of 186 CBCT scans was acquired from two CBCT machines with different acquisition settings. An artificial intelligence (AI) framework was built to segment and classify teeth. Teeth were segmented in a three-step approach with each step consisting of a 3D U-Net and step 2 included classification. The dataset was divided into training set (140 scans) to train the model based on ground-truth segmented teeth, validation set (35 scans) to test the model performance and test set (11 scans) to evaluate the model performance compared to ground-truth. Different evaluation metrics were used such as precision, recall rate and time.

Results

The AI framework correctly segmented teeth with optimal precision (0.98±0.02) and recall (0.83±0.05). The difference between the AI model and ground-truth was 0.56±0.38 mm based on 95% Hausdorff distance confirming the high performance of AI compared to ground-truth. Furthermore, segmentation of all the teeth within a scan was more than 1800 times faster for AI compared to that of an expert. Teeth classification also performed optimally with a recall rate of 98.5% and precision of 97.9%.

Conclusions

The proposed 3D U-Net based AI framework is an accurate and time-efficient deep learning system for automatic tooth segmentation and classification without expert refinement.

Clinical significance

The proposed system might enable potential future applications for diagnostics and treatment planning in the field of digital dentistry, while reducing clinical workload.

Introduction

Tooth segmentation is of vital importance in a daily clinical practice. The identification of teeth with their exact shapes and boundaries on two-dimensional (2D) and three-dimensional (3D) images can guide dental practitioners by allowing an improved precision for early disease detection and diagnosis, treatment planning and outcome prediction [1]. Furthermore, an accurate tooth segmentation for the creation of a 3D tooth model from cone beam computed tomography (CBCT) images is a prerequisite for digital dental workflows [2,3].

An accurate digital model of individual tooth geometry could be beneficial for a number of clinical applications, such as, prosthetic evaluation, orthodontic analysis, orthodontic treatment planning, computer-aided digital implant planning, follow-up of root resorption after orthodontic treatment, canine eruption assessment and tooth auto-transplantation [4], [5], [6], [7]. Additionally, correct tooth detection and segmentation on CBCT images is also crucial for diagnosing pathologies, allowing morphological and positional visualization of teeth to aid the clinical decision-making process [1]. However, an accurate segmentation of individual teeth is an extremely challenging and a time-consuming process.

The conventional image processing techniques for performing tooth segmentation on CBCT images are semi-automated in nature as these require manual intervention and are prone to human error [8]. Similarly, template-based fitting approaches lack robustness for segmenting multi-rooted teeth, and level-set methods need numerous mathematical operations. Furthermore, the vague edges between tooth root and alveolar socket and image intensity inhomogeneity could lead to false segmentation [9]. The aforementioned classical segmentation approaches require laborious manual corrections for achieving an accurate segmentation and are considered as highly time-consuming, operator-dependent and inaccurate especially in the presence of artifacts related to high-density materials [10].

Recently, convolutional neural networks (CNNs) have been widely employed in the field of dentistry for overcoming the limitations associated with the conventional segmentation approaches. Deep neural networks trained end-to-end have the ability to outperform classical pipeline-based systems. These networks have been applied in various fields of image processing, such as, feature extraction, image classification, and semantic segmentation [11]. In context to dentistry, deep learning has allowed detection and segmentation of teeth based on 2D radiography, prediction of third moral eruption, detection and diagnosis of dental caries, and cyst and tumor classification [1,[12], [13], [14], [15], [16]]. However, lack of evidence exists related to the application of deep learning for the segmentation and/or classification of teeth from CBCT images [2,3,10,11,[17], [18], [19], [20], [21]].

A successful tooth segmentation from a clinician's perspective should exhibit the following; accurate segmentation of complete 3D individual teeth, correct classification of each tooth, and fast segmentation and classification [22]. Failure of any of these measures would result in an unsuccessful segmentation task. Additionally, previous evidence also suggests the necessity of further research with more robust, accurate and fast systems, capable of achieving a high segmentation and classification performance for all the teeth groups with images acquired from different devices and protocols [21].

Therefore, the aim of the following study was to develop and validate a clinically operational CNN-based system allowing an accurate and time-efficient segmentation and classification of 3D teeth from CBCT images.

Section snippets

Materials and methods

This study was conducted in compliance with the World Medical Association Declaration of Helsinki on medical research. Ethical approval was obtained from the Ethical Review Board (reference number: S57587). Informed consent was not required for this retrospective study as patient-specific information was kept anonymous.

Results

The timing of segmentation and classification of all the teeth based on the test dataset of a single scan (n = 11 scans with 332 teeth) with the AI model was 13.7 ± 1.2 s compared to that of an expert (25,353.6 ± 4284 s or 7 ± 1.2 h). Thereby, indicating that the AI performed more than 1800 times faster than an expert.

Table 1 describes the accuracy metrics which were calculated for the segmentation evaluation by comparing the AI model to the ground truth. Fig. 3 shows an example of segmentation

Discussion

The 3D visualization and segmentation of human teeth has become an indispensable component for computer aided diagnostics and treatment planning in many fields of digital dentistry. The following study validated a new system for automatic tooth segmentation and classification based on CBCT images acquired by two different acquisition devices with a variety of FOVs and protocol settings. The use of three different CNNs yielded a high accuracy. Furthermore, the AI-driven system performed 1800

Conclusions

This study developed and validated a new cloud-based deep learning system for automatic tooth segmentation and classification without expert refinement.

The proposed system is accurate and time-efficient, enabling potential future applications in the digital workflows of dental diagnostics and treatment planning while reducing clinical workload.

Credit authorship contribution statement

Eman Shaheen: Conceptualization, Methodology, Validation, Formal analysis, Visualization, Writing – original draft. André Leite: Conceptualization, Data curation, Validation, Writing – original draft. Khalid Ayidh Alqahtani: Validation, Data curation, Writing – review & editing. Andreas Smolders: Conceptualization, Methodology, Software, Validation, Writing – original draft. Adriaan Van Gerven: Conceptualization, Methodology, Software, Validation, Formal analysis, Writing – review & editing.

Declaration of Competing Interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (25)

P. Lahoud et al. Artificial intelligence for fast and accurate 3-dimensional tooth segmentation on cone-beam computed tomography J. Endod. (2021)

M. Chung et al. Pose-aware instance segmentation framework from cone beam CT images for tooth segmentation Comput. Biol. Med. (2020)

S.B. Khanagar et al. Developments, application, and performance of artificial intelligence in dentistry – a systematic review J. Dent. Sci. (2021)

M. EzEldeen et al. 3-dimensional analysis of regenerative endodontic treatment outcome J. Endod. (2015)

A.F. Leite et al. Artificial intelligence-driven novel tool for tooth detection and segmentation on panoramic radiographs Clin. Oral Investig. (2021)

Q. Li et al. Automatic tooth roots segmentation of cone beam computed tomography image sequences using U-net and RNN J. Xray. Sci. Technol. (2020)

T.J. Jang et al. A fully automated method for 3D individual tooth identification and segmentation in dental CBCT IEEE Trans. Pattern Anal. Mach. Intell. (2021)

C. Tanikawa et al. Machine/deep learning for performing orthodontic diagnoses and treatment planning
K. Alqahtani et al. Validation of a novel method for canine eruption assessment in unilateral cleft lip and palate patients Clin. Exp. Dent. Res. (2021)