SRMDent: Dental Carries Segmentation using Residual Attention U-Net for Orthopantomogram Images

This study developed an automated deep learning model, the Residual Attention U-Net (RAU-Net), to enhance dental caries diagnosis through improved segmentation. Our approach boosts computing efficiency and feature capture by integrating residual learning and attention mechanisms. Our new dataset, SRMDent, contains 1130 orthopantomogram (OPG) images from SRM Dental College, meticulously analyzed and annotated for cavities. The model was trained using a binary cross-entropy loss function and the Adam optimizer, with early stopping and learning rate reduction to mitigate overfitting. Evaluation metrics, including the Dice and Jaccard coefficients, indicated that the Residual Attention U-Net significantly improved segmentation performance, highlighting its potential to advance dental diagnosis. Additionally, a comparative study was conducted with various U-Net models, including the standard U-Net, Attention U-Net (AU-Net), and Residual Attention U-Net (RAU-Net). The proposed model outperformed the others, demonstrating its effectiveness in accurate and efficient caries detection, with a Mean IoU of 0.1712. For future work, we plan to expand the SRMDent dataset and further explore the integration of 3D imaging to enhance the accuracy and reliability of dental caries detection.