Dental Cavity Analysis in Restorative Dentistry Using Deep Learning and Explainable Artificial Intelligence

In dental education, it is important for instructors to objectively evaluate the cavities prepared by students on mannequin teeth. However, this evaluation process is difficult due to factors such as physical fatigue and eye strain, which can compromise the quality of feedback. Therefore, interest in computer-aided systems that provide objective evaluations is increasing. The rapid advancements in artificial intelligence, particularly deep learning, have shown promise in various fields, including medicine and dentistry. Convolutional neural networks (CNNs), inspired by the mammalian visual system, perform in tasks such as classification and object detection within computer vision. Despite its potential, there is no research on the use of CNN to evaluate cavities. This study aimed to explore the feasibility of using CNNs to classify cavities into narrow, normal width, or wide categories based on photographs. Ten different CNN models were used to classify them. Additionally, the decision-making processes of these models were visualized through heat maps, offering insights into their predictions. According to the test results, the highest accuracy, precision and recall values were found for DenseNet-169 (98.85%, 98.61%, 99.10%). This study can be conceivable for future research in automating cavity evaluations in dental education, enhancing objectivity, and enabling self-assessment for students.