Morphology Prior Enhanced Teeth Segmentation for High-Resolution Oral Scans

Area of Science:

• Medical Imaging
• Computer Vision
• Dental Technology

Background:

• Deep learning for tooth segmentation from intra-oral scans (IOS) is crucial for clinical dental practice.
• Existing methods struggle with low-resolution data, fixed receptive fields, and up-sampling interpolation, leading to inaccurate boundary segmentation.
• Cluttered poses in IOS limit the generalization and usability of geometric information.

Purpose of the Study:

• To develop a morphology prior-enhanced teeth segmentation framework for improved accuracy and generalization.
• To address limitations of existing deep learning methods in handling intra-oral scan data.
• To enhance the adaptability and segmentation of different tooth parts and boundaries.

Main Methods:

• A robust preprocessing step aligns IOS poses using dental arch orientations.
• A decomposition-merging strategy avoids up-sampling limitations by processing low-resolution data and merging results.
• An innovative module integrates semantic and geometric features for adaptive deformable receptive fields.

Main Results:

• The proposed framework significantly outperforms 11 state-of-the-art methods on 6238 IOS from four centers.
• Achieved a 6.93% enhancement in cross-center testing, demonstrating superior generalization.
• The method shows improved adaptability to different tooth parts and accurate segmentation of boundaries.

Conclusions:

• The morphology prior enhanced teeth segmentation framework offers superior performance and generalization for clinical dental applications.
• The novel approach effectively addresses limitations in existing deep learning-based tooth segmentation.
• This work advances the accuracy and reliability of automated dental diagnostics using intra-oral scans.