Evaluating the Precision of Automatic Segmentation of Teeth, Gingiva and Facial Landmarks for 2D Digital Smile Design Using Real-Time Instance Segmentation Network

Abstract

Digital smile design (DSD) technology, which takes pictures of patients’ faces together with anterior dentition and uses them for prosthesis design, has been recently introduced. However, the limitation of DSD is that it evaluates a patient with only one photograph taken in a still state, and the patient’s profile cannot be observed from various viewpoints. Therefore, this study aims to segment the patient’s anterior teeth, gingiva and facial landmarks using YOLACT++. We trained YOLACT++ on the annotated data of the teeth, lips and gingiva from the Flickr-Faces-HQ (FFHQ) data. We evaluated that the model trained by 2D candid facial images for the detection and segmentation of smile characteristics. The results show the possibility of an automated smile characteristic identification system for the automatic and accurate quantitative assessment of a patient’s smile.