Semantic Graph Attention With Explicit Anatomical Association Modeling for Tooth Segmentation From CBCT Images.

Area of Science:

• Radiology
• Medical Imaging
• Computer Vision

Background:

• Accurate tooth identification is crucial for dental diagnosis and treatment planning.
• Existing tooth segmentation methods struggle with similar appearances of adjacent and symmetric teeth.
• The specific anatomical topology of teeth is often overlooked, limiting segmentation accuracy.

Purpose of the Study:

• To develop a semantic graph-based method for precise tooth delineation in dental CBCT images.
• To explicitly model the spatial associations and anatomical topology of teeth.
• To improve segmentation accuracy by addressing challenges like bilateral symmetry confusion.

Main Methods:

• A coarse-to-fine approach using a lightweight network to initially separate teeth into four quadrants.
• A semantic graph attention mechanism to model the anatomical topology within each quadrant.
• Learning voxel-wise discriminative feature embeddings for accurate boundary delineation.

Main Results:

• The proposed semantic graph method significantly enhances tooth segmentation accuracy in dental CBCT images.
• The method effectively handles the confusion between bilaterally symmetric teeth.
• Experimental results show superior performance compared to state-of-the-art segmentation approaches.

Conclusions:

• The semantic graph-based method provides a robust solution for accurate tooth segmentation in dental CBCT.
• Explicitly modeling anatomical topology is key to overcoming limitations of current segmentation techniques.
• This approach holds promise for improving clinical oral diagnosis and treatment planning.