Two-stage deep learning framework for occlusal crown depth image generation.

Area of Science:

• Computer Vision
• Biomedical Imaging
• Dental Technology

Background:

• Generating accurate depth images of dental crowns is challenging due to customization needs.
• Existing computer vision models struggle with dental crowns in fluid positions.
• Reducing dental technician workload is crucial for efficient patient treatment.

Purpose of the Study:

• To propose a two-stage computer vision model for generating depth images of occlusal dental crowns in diverse positions.
• To improve both shape and surface structure accuracy of generated dental crown depth images.
• To enhance the realism and applicability of computer-generated dental crown models.

Main Methods:

• A two-stage model combining segmentation and inpainting networks was developed.
• The segmentation network identifies crown position and size for adaptability.
• A GAN-based inpainting network generates crown surface structures using jaw images and binary masks.

Main Results:

• The proposed model significantly reduced Mean Squared Error (MSE) from 0.007001 to 0.002618.
• The DICE score improved from 0.9333 to 0.9648, indicating better mask accuracy.
• The model demonstrated superior performance in generating realistic crown details compared to baseline methods.

Conclusions:

• The two-stage model effectively generates accurate depth images of dental crowns in varied positions.
• The integration of segmentation and inpainting networks enhances shape and surface structure fidelity.
• This approach offers improved realism and potential for direct use in patient dental treatment.