Akademik Veri Yönetim Sistemi
The 30th IAPD (The International Association of Paediatric Dentistry) Congress , Cape-Town, Güney Afrika, 22 - 25 Ekim 2025, ss.1-2, (Tam Metin Bildiri)
Assesment of Class II Cavity Preparations by Preclinical Students Using Artificial Intelligence: Machine Learning-Based Scoring
Özkuyucu D1, Dayan E1, Araz Topuz M1, Kocaaydın S1, Yağcıoğlu M2, Sepet E1
1Department of Pediatric Dentistry, Faculty of Dentistry, İstanbul Kent University, İstanbul, Türkiye
2Computer Engineering, Faculty of Engineering, Istanbul Arel University, İstanbul, Türkiye
Background/Purpose: Artificial Intelligence (AI) has shown great potential in improving objectivity and standardization across various fields, including dental assessment. This study aims to develop an AI-based model to minimize subjectivity in evaluating student assignments during the preclinical phase of dental education.
Objective: By integrating image processing and machine learning, the model seeks to provide consistent and objective evaluation. The system is expected to reduce observer-related variability and support the adoption of digital, equitable grading practices in dental education.
Methods: The study includes 250 artificial primary molar samples with Class II cavity preparations designed according to Black’s cavity principles. To enable objective measurement of the preparations, routine evaluation criteria were adapted for AI-assisted analysis and categorized under ten headings. These parameters included outline form, retention form, surface smoothness, depth control, proximal box width, gingival floor depth, axial wall alignment, marginal ridge thickness, dovetail cavity and the presence of critical errors. Each preparation was scored by three independent instructors based on predefined criteria. Each specimen was photographed from occlusal and proximal perspectives using a DSLR camera (Nikon D7200) under standardized conditions to ensure spatial consistency and image quality. A dual-view Convolutional Neural Network (CNN) architecture was developed to extract and integrate features from paired images. The model was trained on 80% of the labeled dataset and validated on the remaining 20%. Model performance was quantitatively assessed using three standard regression metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE).
Results: On the held-out validation set, the model achieved an MSE of 36.88, an MAE of 4.45, and an RMSE of 6.07 demonstrating strong predictive accuracy.
Conclusion: The results show strong agreement between the model’s predictions and expert scores, suggesting that the CNN model can effectively replicate human-level accuracy in evaluating dental cavity preparations.