Study on virtual tooth image generation utilizing CF-fill and Pix2pix for data augmentation.

Area of Science:

• Artificial Intelligence in Dentistry
• Digital Dentistry
• Medical Imaging

Background:

• Traditional dental prosthetics are labor-intensive and time-consuming.
• Deep learning models can automate dental prosthetic processes but require extensive training data.
• Patient privacy concerns limit the use of real dental images in research.

Purpose of the Study:

• To develop a method for generating realistic virtual tooth images for training AI models.
• To address the challenge of limited access to real patient dental data.

Main Methods:

• Utilized image-to-image translation (pix2pix) to generate diverse virtual tooth images.
• Employed contextual reconstruction fill (CR-Fill) to refine virtual images and ensure anatomical accuracy.
• Compared the performance of deep learning models trained on real images, virtual images, and a combination of both.

Main Results:

• The proposed method successfully generated virtual tooth images visually similar to real intraoral scans.
• Models trained solely on virtual images showed suboptimal performance.
• Training models with a combination of real and virtual images achieved performance comparable to models trained exclusively on real images.

Conclusions:

• AI-driven generation of virtual dental images is a viable solution to overcome data limitations in research.
• Hybrid training datasets (real and virtual images) offer a promising approach for developing robust AI models in dental prosthetics.
• This method can accelerate the development and adoption of AI in digital dentistry.