Multi-Regional deep learning models for identifying dental restorations and prosthesis in panoramic radiographs

ABSTRACT

Background

This study introduces a novel deep learning methodology for the automated detection of a wide range of dental prostheses, including crowns, bridges, and implants, as well as various dental treatments such as fillings, root canal therapies, and endodontic posts.

Methods

Critically, our model was trained and validated using a diverse, multi-regional population dataset comprising 2,235 panoramic radiographs sourced from three distinct dental colleges. This diverse dataset enhances the generalizability and robustness of our approach. Our rigorous workflow encompassed dataset preparation, deep learning model selection, comprehensive training and evaluation procedures, and detailed result analysis and interpretation. We evaluated the performance of three state-of-the-art deep learning architectures: You Only Look Once (YOLO)v11, Faster Region-based Convolutional Neural Network (Faster R-CNN), and Vision Transformer (ViT). Each model was rigorously tested using a standardized protocol involving the division of the dataset into training, validation, and test sets.

Results

Our results demonstrate the superior performance of the ViT model, achieving a remarkable detection accuracy of 94.15%, coupled with a high precision of 94.64% and an F1-score of 93.52%. Interestingly, while ViT excelled in these specific metrics, Faster R-CNN yielded the best mean average precision (mAP) of 82.0% at an Intersection over Union (IoU) threshold of 0.50. This comparative analysis provides valuable insights into the strengths and weaknesses of different deep learning architectures for this specific task.

Conclusion

This research provides compelling evidence supporting the practicality and effectiveness of AI-driven automated dental restoration identification from panoramic radiographs, offering a highly promising and efficient solution for enhancing clinical diagnostic accuracy and streamlining workflows in dental practice. This AI, achieving 94.15% accuracy on diverse panoramic radiographs, automates the detection of numerous dental restorations. It provides compelling evidence for AI’s practicality in significantly enhancing diagnostic precision and streamlining clinical workflows, offering a highly efficient tool for modern dental practice and improved patient record management.