AI Tooth Segmentation in Digital Dental Models Using Deep Learning

Accuracy and Efficiency of Automatic Tooth Segmentation in Digital Dental Models Using Deep Learning

Authors:

Joon Im; Ju‑Yeong Kim; Hyung‑Seog Yu; Kee‑Joon Lee; Sung‑Hwan Choi; Ji‑Hoi Kim; Hee‑Kap Ahn; Jung‑Yul Cha

Summary:

This study evaluates a dynamic graph convolutional neural network (DGCNN)-based method for automatic tooth segmentation and classification in 3D digital dental models. Using 516 models for training and 30 for evaluation, the authors compared three segmentation approaches: automatic segmentation (AS, deep learning), landmark‑based segmentation (LS), and tooth designation segmentation (DS). Assessments included success rate, mesiodistal (MD) width, clinical crown height (CCH), and segmentation time. AS achieved the shortest segmentation time (57.73 s) and high success rate (97.26%). Deep learning-based segmentation demonstrated high accuracy and efficiency suitable for orthodontic diagnosis and appliance fabrication.

Key Words:

3D Models; Tooth Segmentation; DGCNN; Point Cloud; Orthodontics; Deep Learning

Extracted Data:

Year:

2022

Modality:

3D Dental Models (STL/Point Cloud)

Dataset:

516 models for training; 30 models for evaluation

Dataset Split:

Training set (516) / Evaluation set (30 models, 840 teeth)

Network Architecture:

Two‑stage DGCNN (binary + 17‑class) + curvature‑based mesh refinement

Metrics:

Success Rate (AS: 97.26%); MD width/CCH error; Segmentation Time (AS: 57.73 s)

AP - Professional Qty:

Orthodontic specialist + evaluation by multiple experts

AP - Supervisor Presence:

Yes

AP - Experience Level:

Specialist-level

AP - Expertise Area:

Orthodontics

AP - Tool or System:

LaonSetup (AS); OrthoAnalyzer (LS); Autolign (DS); Meshmixer; Geomagic Control X

ML Task:

3D Semantic Segmentation + Tooth Classification

Project Objective:

Compare deep learning-based segmentation against conventional CAD/CAM segmentation tools.

Clinical Relevance:

High-accuracy automated segmentation improves orthodontic diagnosis, model setup, and appliance fabrication, reducing manual labor and increasing workflow efficiency.