Abstract
Panoramic radiography is the most popular imaging modality in dentistry. On the other hand, cone-beam computed tomography (CBCT) is usually utilized to review three-dimensional dental structures in high resolution, however, it is less utilized in dental clinics due to its high cost and radiation exposure. An automated method for reconstructing 3D oral cavity structure from panoramic X-ray will be beneficial for patients to reduce expenses for dental care without radiation exposure. As an initial step to the 3D shape reconstruction of the oral cavity, a flattened CBCT 3D volume across the dental arch can be generated from a 2D panoramic X-ray image using a deep learning method. Existing deep learning models are almost based on the encoder-decoder architecture, but their performance has not been well elucidated. In this paper, we evaluated the applicability of attention U-net for reconstructing the flattened CBCT 3D volume from a panoramic X-ray. In our quantitative and qualitative evaluations, the attention U-net outperformed the auto-encoder and led to enhanced perceptual quality.
Funding Information
  • Ministry of SMEs and Startups (S3228660)
  • National Research Foundation of Korea (NRF) (2021RIS-004)

This publication has 12 references indexed in Scilit: