Can super resolution via deep learning improve classification accuracy in dental radiography?

Area of Science:

• Artificial Intelligence
• Medical Imaging

Background:

• Deep Learning (DL) based Super Resolution (SR) enhances image quality.
• SR's impact on dental image classification is under-researched.
• This study evaluates SR's effect on DL dental classification performance.

Purpose of the Study:

• To assess the performance of DL classification models on dental images enhanced by SR.
• To compare classification results with and without SR pre-processing.
• To investigate the impact of different SR models and scaling ratios on classification outcomes.

Main Methods:

• Utilized an open-source dental image dataset.
• Applied two SR models with scaling ratios of 2x and 4x.
• Evaluated classification performance using four DL models and metrics like accuracy, F1-score, SSIM, and PSNR.

Main Results:

• SR models achieved high image quality metrics (SSIM: 0.904, PSNR: 36.71).
• Classification using SR images yielded average accuracy of 0.859 and F1-score of 0.873.
• Two comparison approaches showed improved classification in 50% to 75% of cases with SR.

Conclusions:

• SR-generated images significantly enhance dental image classification performance.
• This is the first study to investigate SR for improved resolution in dental radiographs for classification.
• SR offers a promising approach to boost diagnostic accuracy in AI-driven dental imaging.