Development and Evaluation of a Deep Learning Model for Occlusion Classification in Intraoral Photographs

Occlusal classification is crucial for designing orthodontic treatment plans. Therefore, this study aims to develop a deep learning tool to identify occlusal types from digital oral photographs automatically. Using a large-scale dataset with high-quality annotations (including 6100 orthodontic intraoral pictures at a 45° lateral view), two deep-learning models were developed based on ResNet-18 to identify various occlusal classifications: molar occlusal relationships (M1, M2, M3) and canine occlusal relationships (C1, C2, C3). Additionally, we developed models based on ConvNeXt and SVM for comparison. Our model achieves an accuracy of 0.90 and a weighted average F1-score of 0.90 in identifying various molar occlusal relationships. For identifying different canine occlusal relationships, the model attains an accuracy of 0.89 and a weighted average F1-score of 0.89. The performance of our model demonstrates its potential for clinical applications.