TAPSeg: An open-source deep learning tool for instance-level tooth and pulp segmentation in CBCT

Area of Science:

• * Dental imaging and computational anatomy.
• * Deep learning applications in medical image analysis.

Background:

• * Cone-beam computed tomography (CBCT) is crucial for dental diagnostics.
• * Accurate segmentation of teeth and pulp is essential for quantitative analysis and treatment planning.
• * Current segmentation methods can be time-consuming and require specialized expertise.

Purpose of the Study:

• * To develop an open-source, one-tap tool for automatic tooth and pulp segmentation using deep learning.
• * To integrate this tool into the 3D Slicer software for seamless workflow.
• * To validate the tool's generalization capabilities across diverse datasets, including immature teeth.

Main Methods:

• * Developed a three-stage V-Net framework for tooth segmentation (arch positioning, centroid detection, instance segmentation).
• * Utilized nnU-Net for 3D semantic segmentation of the pulp.
• * Evaluated performance using Dice Similarity Coefficient (DSC), Hausdorff Distance (HD95), sensitivity, and precision on multiple datasets.

Main Results:

• * TAPSeg demonstrated robust performance for tooth (DSC: 91.5%-94.2%) and pulp (DSC: 91.0%-92.2%) segmentation.
• * Achieved low HD95 values (0.700-1.449 mm for teeth, 0.704-1.008 mm for pulp).
• * Hard-tissue segmentation for immature teeth yielded a DSC of 92.1% ± 6.5%.

Conclusions:

• * TAPSeg, integrating V-Net and nnU-Net, offers accurate, efficient, and generalizable tooth-pulp segmentation.
• * The one-click 3D Slicer plug-in simplifies clinical use and lowers the technical threshold.
• * Enables rapid generation of patient-specific 3D models for quantitative analysis and advanced dental procedures.