A self attention based deep learning framework for accurate and efficient dental disease detection in OPG radiographs

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Oral diseases are a growing concern, necessitating accurate and efficient diagnostic tools.
• Current diagnostic methods can be prone to human error and require significant clinician time.
• Advancements in deep learning offer potential for automated diagnostic systems in dentistry.

Purpose of the Study:

• To propose and compare deep learning models, specifically Vision Transformer (ViT) and Swin Transformer, for the automated diagnosis of teeth diseases using Orthopantomogram (OPG) images.
• To evaluate the performance of ViT and Swin Transformer in terms of accuracy, precision, and recall for oral disease detection.
• To assess the efficiency and scalability of transformer-based architectures in dental imaging.

Main Methods:

• Utilized deep learning techniques, focusing on transformer-based architectures: Vision Transformer (ViT) and Swin Transformer.
• Trained and evaluated the models on Orthopantomogram (OPG) images for teeth disease diagnosis.
• Compared the performance metrics, including test accuracy, precision, and recall, between ViT and Swin Transformer.

Main Results:

• Vision Transformer (ViT) achieved a test accuracy of 96%, precision of 95.8%, and recall of 96.2%.
• Swin Transformer demonstrated a competitive accuracy of 95.2% with efficient inference time and scalable complexity.
• ViT exhibited superior performance compared to Swin Transformer in diagnosing oral diseases.

Conclusions:

• Transformer-based architectures, particularly ViT, are highly effective for automated diagnosis in dental imaging.
• The proposed approach offers a promising solution for clinicians, enhancing accuracy and reducing diagnosis time for oral diseases.
• Deep learning models can significantly improve the detection of oral diseases, addressing the increasing demand for advanced diagnostic systems.