Synthetic Orthopantomography Image Generation Using Generative Adversarial Networks for Data Augmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Annotated dental X-ray datasets are scarce, hindering AI development.
• Existing datasets pose privacy and resource limitations.

Purpose of the Study:

• To develop a novel pipeline for generating high-resolution synthetic orthopantomography (OPG) images.
• To address the challenge of limited annotated dental X-ray data for AI applications.

Main Methods:

• Utilized customized generative adversarial networks (GANs) trained on 4777 real OPG images.
• Generated synthetic OPGs at 2048 × 1024 resolution, preserving fine anatomical detail.
• Evaluated synthetic images using a YOLO object detection model and expert dentist scoring.

Main Results:

• Selected GAN models produced realistic synthetic OPGs with accurate structural representation.
• YOLO detector demonstrated strong performance on synthetic images, indicating reliable feature representation.
• Expert evaluations confirmed high anatomical plausibility, with top models achieving over 50% of real OPG scores.

Conclusions:

• The GAN-based pipeline enables ethical and scalable creation of synthetic OPG images.
• This method effectively augments datasets for AI-driven dental diagnostics.
• Provides a practical solution for AI model development in data-limited or privacy-sensitive settings.