A Tooth Segmentation Method Based on Multiple Geometric Feature Learning.

Area of Science:

• Computer Vision
• Medical Imaging
• Orthodontics

Background:

• Accurate tooth segmentation is crucial for virtual orthodontic systems.
• Existing deep learning methods struggle to effectively differentiate point coordinate and normal vector information, hindering feature complementarity.

Purpose of the Study:

• To propose a novel tooth segmentation method utilizing multiple geometric feature learning.
• To improve the accuracy and efficiency of tooth segmentation in dental models.

Main Methods:

• Employed a spatial transformation (T-Net) module for aligning dental model mesh features.
• Developed a multiple geometric feature learning module to encode and enhance centroid coordinates and normal vectors.
• Implemented feature downscaling and channel optimization using multilayer perceptron (MLP) and efficient channel attention (ECA) for local to global feature fusion.

Main Results:

• The proposed algorithm demonstrated superior accuracy and efficiency in tooth segmentation compared to existing methods.
• The enhanced feature learning effectively highlighted differences between geometric features of various tooth meshes.

Conclusions:

• The novel method significantly improves tooth segmentation in virtual orthodontic applications.
• This technique can provide valuable assistance to dentists in treatment planning and execution.