TSRNet: A Dual-Stream Network for Refining 3D Tooth Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Deep learning has advanced 3D tooth segmentation.
• Existing methods struggle with imprecise segmentation boundaries and prediction inaccuracies.
• Refining coarse segmentation results is crucial for clinical applications.

Purpose of the Study:

• To introduce a novel, learnable method for refining coarse 3D tooth segmentation results.
• To address challenges in boundary accuracy and prediction errors in current 3D tooth segmentation algorithms.
• To improve the precision of 3D tooth segmentation for better downstream analysis.

Main Methods:

• Developed a dual-stream network, TSRNet (Tooth Segmentation Refinement Network), for segmentation refinement.
• Utilized explicit boundary maps and gradient information from distance maps for rectification.
• Employed an iterative refinement process guided by refined boundary and distance maps.

Main Results:

• TSRNet effectively rectifies defective boundary and distance maps from coarse segmentations.
• The two-stage refinement method demonstrated significant improvements on benchmark datasets.
• Achieved state-of-the-art performance in 3D tooth segmentation refinement.

Conclusions:

• The proposed TSRNet offers a significant advancement in refining 3D tooth segmentation.
• The method successfully improves upon baseline coarse segmentation results.
• This approach holds promise for enhancing the accuracy of automated dental analysis.