TSegLab: Multi-stage 3D dental scan segmentation and labeling.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate 3D teeth segmentation and labeling are crucial for dental computer-aided design (CAD) applications.
• Existing methods often struggle with precision and reliability in processing complex 3D dental scans.
• Deep learning offers potential for automated and enhanced analysis of dental structures.

Purpose of the Study:

• To introduce a novel deep learning framework for accurate 3D teeth scan segmentation and labeling.
• To improve the precision and reliability of automated dental analysis within CAD systems.
• To establish a new state-of-the-art in teeth localization, segmentation, and identification.

Main Methods:

• A three-stage deep learning pipeline: coarse localization using Mask-RCNN on 2D representations.
• Fine teeth segmentation via Mask-RCNN on isomorphically mapped 2D harmonic parameter spaces.
• Labeling using a graph neural network incorporating 3D shape and spatial distribution, with novel data augmentation for variations.

Main Results:

• Achieved high performance on the Teeth3DS dataset (1800 scans): Teeth Localization Accuracy (TLA) of 98.45%, Teeth Segmentation Accuracy (TSA) of 98.17%, and Teeth Identification Rate (TIR) of 97.61%.
• Outperformed existing state-of-the-art techniques in all evaluated metrics.
• Demonstrated significant enhancement in the precision and reliability of automated teeth segmentation and labeling.

Conclusions:

• The proposed deep learning approach significantly advances automated 3D teeth segmentation and labeling.
• The method's high accuracy and reliability make it a valuable tool for dental CAD applications.
• This work paves the way for more sophisticated AI-driven solutions in digital dentistry.