Faculty of Dental Science UOP

As artificial intelligence (AI) transforms medical diagnostics and training, dental education stands poised for its own revolution, where machine learning can match expert judgment in evaluating clinical precision. Manual assessment of cavity preparations in dental education is still inflexible, time-consuming, and largely dependent on faculty availability issues, which are exacerbated in settings with limited resources. This work presents a deep learning-based grading and feedback system that uses top-view clinical images of student-prepared cavities. The proposed method employs a MobileNetV2-based Convolutional Neural Network (CNN), which is optimized via transfer learning and extensive augmentation of a limited expert-annotated dataset of 137 samples. The system independently evaluates four clinically important characteristics that dental educators define: smooth outline, flat floor, depth, and undercut. Across the four categories, experimental evaluation results show test accuracies of 67.74%, 65.32%, 51.61%, and 34.67%. Faculty and student qualitative validation confirms the pedagogical relevance and alignment with the academic grading criteria of the system. With the potential to improve feedback quality and instructional efficiency, this work presents a unique and scalable artificial intelligence-driven framework for automated, consistent, and timely evaluation in dental education.