DilatedToothSegNet: Tooth Segmentation Network on 3D Dental Meshes Through Increasing Receptive Vision.

Area of Science:

• Computer-aided dentistry
• Geometric deep learning
• 3D surface segmentation

Background:

• 3D dental models from intraoral scanners are vital for computer-aided treatment planning.
• Manual segmentation and labeling of dental models are time-consuming and prone to errors.
• Existing deep learning methods struggle with atypical dental anatomy, missing teeth, or severe misalignment.

Purpose of the Study:

• To develop an automated method for accurate segmentation of 3D dental surface models.
• To address limitations in current deep learning approaches for challenging dental cases.
• To improve the efficiency and precision of dental model processing for orthodontic and dental applications.

Main Methods:

• Introduction of a novel network operator: dilated edge convolution.
• Expansion of the network's receptive field to capture more distant features.
• Extensive evaluation using the Teeth3DS benchmark dataset for quantitative and qualitative analysis.

Main Results:

• Dilated edge convolution significantly enhances the network's ability to learn complex features.
• Improved segmentation accuracy, especially in cases with missing or misaligned teeth.
• Demonstrated superiority over state-of-the-art methods in quantitative and qualitative assessments.

Conclusions:

• The proposed dilated edge convolution method offers superior performance for 3D dental surface segmentation.
• This advancement can lead to more efficient and accurate computer-aided treatment planning in dentistry and orthodontics.
• The method shows promise for handling complex and diverse patient-specific dental data.