Merging and Annotating Teeth and Roots from Automated Segmentation of Multimodal Images

• Published in: Multimodal Learning for Clinical Decision Support Vol. 13050; pp. 81 - 92
• Main Authors: Deleat-Besson, Romain, Le, Celia, Zhang, Winston, Turkestani, Najla Al, Cevidanes, Lucia, Bianchi, Jonas, Ruellas, Antonio, Gurgel, Marcela, Massaro, Camila, Del Castillo, Aron Aliaga, Ioshida, Marcos, Yatabe, Marilia, Benavides, Erika, Rios, Hector, Soki, Fabiana, Neiva, Gisele, Najarian, Kayvan, Gryak, Jonathan, Styner, Martin, Aristizabal, Juan Fernando, Rey, Diego, Alvarez, Maria Antonia, Bert, Loris, Soroushmehr, Reza, Prieto, Juan
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2021; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Deep learning; Dental crown segmentation; Dentistry; Merging; Root canal segmentation; Universal label
• DOI: 10.1007/978-3-030-89847-2_8
• ISBN: 9783030898465; 3030898466
• ISSN: 0302-9743

This paper aims to combine two different imaging techniques to create an accurate 3D model representation of root canals and dental crowns. We combine Cone-Beam Computed Tomography (CBCT) (root canals) and Intra Oral Scans (IOS) (dental crowns). The Root Canal Segmentation algorithm relies on a U-Net architecture with 2D sliced images from CBCT scans as its input. The segmentation task achieved an F1-score of 0.84. The IOS segmentation (Dental Model Segmentation) algorithm and Universal Labeling and Merging (ULM) algorithm use a multi-view approach for 3D shape analysis. The approach consists of acquiring views of the 3D object from different viewpoints and extract surface features such as the normal vectors. The generated 2D images are then analyzed via a 2D convolutional neural networks (U-Net) for segmentation or classification tasks. The segmentation task on IOS achieved an accuracy of 0.9. The ULM algorithm classifies the jaws between upper and lower and aligns them to a template and labels each crown and root with the ‘Universal Numbering System’ proposed by the ‘American Dental Association’. The ULM task achieve an F1-score of 0.85. Merging and annotation of CBCT and IOS imaging modalities will help guide clinical decision support and quantitative treatment planning for specific teeth, implant placement, root canal treatment, restorative procedures, or biomechanics of tooth movement in orthodontics.