A Reliable Multi-Stage System for Tooth Instance Segmentation and Numbering in Panoramic Radiographs.

Area of Science:

• Medical Imaging
• Computer Vision
• Dental Informatics

Background:

• Accurate tooth segmentation and identification are crucial for automated dental diagnosis.
• Current methods struggle with anatomical symmetry and anomalies like supernumerary or missing teeth.

Purpose of the Study:

• To develop a reliable multi-stage framework for tooth instance segmentation and FDI-compliant numbering.
• To address limitations of existing methods in handling anatomical variations and anomalies.

Main Methods:

• Proposed a multi-stage framework including a supernumerary tooth classifier, core instance segmentor, and missing tooth detector.
• Introduced the Symmetry-Aware Dual-branch Pyramid Network (SADP-Net) with Symmetry-Aware Module (SymAM) and Dual-Branch Pyramid (DBP).

Main Results:

• Demonstrated superior performance over state-of-the-art baselines on three diverse datasets.
• Ablation studies confirmed the effectiveness of SADP-Net components in improving localization, numbering, and robustness.

Conclusions:

• The proposed framework offers a scalable and interpretable solution for clinical dental imaging.
• The SADP-Net effectively handles tooth instance segmentation and numbering, including anomalies.