Deep Learning For Caries Severity Classification On CBCT Images

This study aimed to develop a deep learning algorithm, CariesAI-3D, to achieve accurate classification of caries severity on CBCT images. A high-quality CBCT dataset with 2,148 well-annotated single-tooth images was established and divided into a five-fold cross-validation set (1,826) and an independent test set (322). We developed CariesAI-3D as a multi-task learning network with a Spatial-Attention Feature Fusion Module (SA-FFM) for caries classification. The performance of CariesAI-3D was evaluated in comparison with four baseline models using cross-validation. An ablation study was conducted to validate the effectiveness of the SA-FFM. Performance metrics included accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic (ROC) curve (AUC). CariesAI-3D outperformed the four baseline models on the cross-validation set, achieving an accuracy of 88.61%, precision of 88.20%, recall of 87.26%, and an F1-score of 87.55%. The ablation study demonstrated that the CariesAI-3D with a SA-FFM was more accurate than both the backbone model and the model employing an element-wise feature addition. Furthermore, CariesAI-3D demonstrated good generalisation on the independent test set, as evidenced by a F1-score of 86.50%. By integrating multi-task learning with a SA-FFM, the CariesAI-3D achieved accurate and robust classification of caries severity on CBCT images, demonstrating significant advancements over traditional methods.