Artificial Intelligence Designed Single Tooth Dental Prostheses

Overview

Tooth loss is common and as consequence deteriorate patient's health and quality-of-life. Dental prostheses aim to restore patients' appearance and functions by replacement of missing teeth. The occlusal morphology and 3D position of the healthy natural teeth should be adopted by the dental prostheses (biomimetic). Despite computer-assisted design (CAD) software are available for designing dental prostheses, considerable clinical time are still required to fit the dental prostheses into patients' occlusion (teeth-to-teeth relationship). Teeth of an individual subjects are genetically controlled and exposed to mostly identical oral environment, therefore the occlusal morphology and 3D position of teeth are inter-related. It is hypothesized that artificial intelligence (AI) can automated designing the single-tooth dental prostheses from the features of remaining dentition.

Objectives: 1. To compare four deep-learning methods/algorithms in interpreting and learning of the features of 3D models; 2. To compare the AI system with maxillary tooth model alone to maxillary and mandibular (antagonist) models; 3. To compare the occlusal morphology and 3D position of the single-tooth dental prostheses designed by trained AI and by dental technicians. Methods: First, investigators will collect 200 maxillary dentate teeth models as training models. AI will learn the relationship between individual teeth and rest of the dentition using the 3D Generative Adversarial Network (GAN) by following deep-learning methods/algorithms: Group 1) Voxel-based; Group 2) View-based; Group 3) Point-based; and Group 4) Fusion methods. Investigators will collect another 100 maxillary models that serve as validation models. Investigators will remove a tooth (act as control) in each model. Then investigators will evaluate these deep learning algorithms in predicting the occlusal morphology and 3D position of single-missing tooth. Second, investigators will evaluate the need of antagonist model in predicting the occlusal morphology and 3D position of single-missing tooth in 100 validation models: Group i) maxillary model only and Group ii) with antagonist model using the tested deep-learning algorithm in objective (1). Third, investigators will analyze the geometric morphometric and 3D position of dental prostheses designed by: Group a) the trained AI system; Group b) dental technicians on the physical models; and Group c) dental technicians using CAD software. Investigators will compare these teeth to the corresponding natural teeth (control) in 100 validation models. Furthermore, investigators will analyze the time required for tooth design in these groups as secondary outcome.