Fully Automated Tooth Segmentation and Labeling for Both Full- and Partial-Arch Intraoral Scans Using Deep Learning.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Digital Dentistry

Background:

• Partial-arch intraoral scans (IOSs) are crucial in clinical dentistry but pose challenges for existing tooth segmentation algorithms.
• Current algorithms often perform poorly on partial-arch data, limiting their clinical applicability.

Purpose of the Study:

• To develop a fully automated deep learning (DL) model for precise tooth segmentation and labeling on both full- and partial-arch IOSs.
• To enhance the accuracy and reliability of AI in analyzing diverse intraoral scan data.

Main Methods:

• A two-stage DL model (ToothInstanceNet) was developed, incorporating artificial partial-arch IOSs, a DL-based alignment module, and FDI-aware postprocessing.
• The model was trained on 600 IOSs (300 full-arch, 300 partial-arch) and validated using 5-fold cross-validation and the public Teeth3DS dataset.

Main Results:

• The model achieved high performance metrics, including F1-scores of 0.9908 (full-arch) and 0.9884 (partial-arch), and Dice scores of 0.9819 and 0.9862, respectively.
• Superior performance was demonstrated in the 3DTeethSeg challenge (score = 0.9870).
• Analysis revealed correlations between model errors and specific dental conditions like residual roots and missing teeth.

Conclusions:

• The study presents the first fully automated method for tooth segmentation and FDI labeling applicable to both full- and partial-arch IOSs.
• The model's high accuracy indicates its potential for seamless integration into clinical dental workflows.
• This advancement facilitates tooth identification and supports automated downstream applications in digital dentistry.