A fine-grained orthodontics segmentation model for 3D intraoral scan data

Area of Science:

• Computer Vision
• Medical Imaging
• Dental Technology

Background:

• Digital orthodontics relies on accurate tooth segmentation from 3D intraoral scans.
• Existing datasets often use indirect scans and lack diverse, clinically relevant samples, hindering real-world application.
• Accurate segmentation is crucial for subsequent dental diagnosis and treatment planning.

Purpose of the Study:

• To address the lack of standardized datasets for tooth segmentation analysis.
• To introduce a fine-grained 3D intraoral scan dataset (3D-IOSSeg) for deformed teeth segmentation.
• To propose and validate a novel fast graph convolutional network (Fast-TGCN) for accurate 3D tooth segmentation.

Main Methods:

• Development of the 3D-IOSSeg dataset with over 200 patients' 3D intraoral scan data, featuring fine-grained mesh unit labels for all teeth.
• Proposal of Fast-TGCN, a graph convolutional network utilizing a naive adjacency matrix to capture local geometric features.
• Extensive experimentation comparing Fast-TGCN against other segmentation methods on the 3D-IOSSeg dataset.

Main Results:

• Fast-TGCN demonstrated superior speed and accuracy in segmenting teeth from complex intraoral scans.
• The proposed method outperformed existing approaches across various evaluation metrics.
• Comprehensive analysis of classical tooth segmentation methods on the new dataset was provided.

Conclusions:

• The 3D-IOSSeg dataset offers a standardized resource for evaluating tooth segmentation algorithms, particularly for deformed teeth.
• Fast-TGCN presents an effective and efficient solution for 3D tooth segmentation in clinical orthodontic scenarios.
• The study advances the field of digital orthodontics by improving the accuracy and applicability of automated tooth segmentation.