Super-Resolution of Dental Panoramic Radiographs Using Deep Learning: A Pilot Study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Super-resolution (SR) algorithms can upscale low-resolution images to high-quality ones.
• Improving the resolution of dental panoramic radiographs is crucial for accurate diagnosis.

Purpose of the Study:

• To compare the performance of deep learning-based SR models against a conventional approach for enhancing dental panoramic radiographs.
• To evaluate state-of-the-art deep learning SR models including SRCNN, SRGAN, U-Net, SwinIr, and LTE.

Main Methods:

• A dataset of 888 dental panoramic radiographs was used.
• Five deep learning SR models (SRCNN, SRGAN, U-Net, SwinIr, LTE) and bicubic interpolation were evaluated.
• Performance metrics included Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Mean Opinion Score (MOS) by experts.

Main Results:

• The local texture estimator (LTE) model demonstrated the highest performance.
• LTE achieved the best scores for MSE (7.42 ± 0.44), SSIM (0.919 ± 0.003), PSNR (39.74 ± 0.17), and MOS (3.59 ± 0.54).
• All tested SR approaches showed significant MOS improvements compared to low-resolution images.

Conclusions:

• Super-resolution significantly enhances the quality of dental panoramic radiographs.
• The LTE deep learning model is the most effective among the evaluated methods for improving radiograph resolution.