Artificial Intelligence Precision Recognition and Auxiliary Diagnosis of Dental X-ray Panoramic Images Based on Deep Learning

¹ School of Stomatology, Hangzhou Normal University, Hangzhou 311121, Zhejiang, China
² Liushi Oral Hospital, Liushi, 323000, Zhejiang, China

^* Corresponding author: c18357757158@126.com

Abstract

Objective: This study aims to explore the application of deep learning algorithms in dental X-ray panoramic images, particularly for the automatic segmentation of dental caries and identification of wisdom tooth types, in order to improve the accuracy and efficiency of dental diagnosis and assist doctors in formulating precise treatment plans. Methods: Multiple classic medical image segmentation network models (including Unet, PSPNet, FPN, Unet++, and DeepLabV3+) were trained and tested on the ParaDentCaries dataset to evaluate their performance in dental X-ray panoramic images. Performance was comprehensively compared using evaluation metrics such as IoU (Intersection over Union), Dice coefficient, sensitivity, accuracy, Hd95 (Hausdorff distance), and model parameters. Additionally, visualization methods were used to display the model’s prediction results across different lesion scales (small caries, medium caries, large caries). Results: The experimental results show that the Unet++ model performed best across all evaluation metrics, especially achieving strong results in IoU (58.94), Dice coefficient (72.71), sensitivity (68.03), and accuracy (82.00). Compared to other models (such as FPN, PSPNet, DeepLabV3+), Unet++ demonstrated clear advantages in detail recognition and boundary handling, particularly exhibiting higher accuracy in detecting small and medium caries. Visual analysis showed that Unet++ was able to accurately identify secondary carious areas, while PSPNet and DeepLabV3+ performed poorly in this regard, showing boundary detection deviations. Conclusion: The deep learning-based automatic diagnostic system for dental X-ray panoramic images, especially the Unet++ model, provides high-precision predictive results in dental caries segmentation and wisdom tooth type identification, significantly improving diagnostic efficiency and accuracy.