Artificial Intelligence-drive tool for automatic segmentation of endodontic structures in cone-beam computed tomography images

• Grant number: 22/13774-3
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: March 01, 2023
• End date: September 30, 2023
• Field of knowledge: Health Sciences - Dentistry - Endodontics
• Principal Investigator: Mario Tanomaru Filho
• Grantee: Airton Oliveira Santos Junior
• Supervisor: Reinhilde Jacobs
• Host Institution: Faculdade de Odontologia (FOAr). Universidade Estadual Paulista (UNESP). Campus de Araraquara. Araraquara , SP, Brazil
• Institution abroad: University of Leuven, Leuven (KU Leuven), Belgium
• Associated to the scholarship: 20/11012-3 - Comparative evaluation between micro-CT and CBCT using the e-Vol DX software in the analysis of obturation in oval canals and dentinal thickness in MB1 and MB2 canals of maxillary molars after preparation and retreatment, BP.DR

Abstract

The segmentation of dentomaxillofacial structures in cone beam computed tomography (CBCT) imaging is a challenging task, especially due to limited contrast resolution, presence of several types of artifacts, as well as the partial volume effect in CBCT images. Thus, artificial intelligence (AI)-driven algorithms have been proposed for automated segmentation of anatomical structures and teeth. The aims of this study will be to train an AI-driven algorithm, composed of two convolutional neural networks in deep and specific learning, for automatic segmentation of endodontic structures (root canal and pulp chamber) using CBCT exams acquired in high resolution protocol; and then, to validate the accuracy of this Al-driven algorithm by comparing it to human performance. A total of 175 CBCT images of single and double-rooted teeth will be anonymously selected and randomly divided to the training stages (140 exams) and validation (35 exams). The images will be manually segmented by an Endodontic Specialist using MeVisLab software (MeVis Research, Bremen, Germany), and the dataset will be virtually imported for training the AI algorithm. The segmentation of endodontic structures will be performed using three different protocols: manual segmentation performed by a Specialists, automatic segmentation guided by the AI algorithm and automated segmentation performed by AI and manual refined by Specialists. The time required to execute each of the protocols will also be evaluated. All stages of this project will be performed in a partnership proposed with the Katholieke Universiteit Leuven, KU Leuven, Belgium (OMFS IMPATH Research Group). The results obtained will be submitted to the proper statistical tests with a significance level of 5%. (AU)