Evaluation of Apical Closure in Panoramic Radiographs Using Vision Transformer Architectures ViT-Based Apical Closure...

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Open apex is a developmental anomaly affecting tooth maturation.
• Accurate classification of open apex on panoramic radiographs (OPGs) is crucial for treatment planning.
• Deep learning models offer potential for automated analysis of dental radiographs.

Purpose of the Study:

• To evaluate vision transformer (ViT)-based deep learning models for open apex classification on OPGs.
• To compare the diagnostic accuracy of ViT models against conventional convolutional neural network (CNN) architectures.
• To assess the performance of various classifiers when used with ViT and CNN models.

Main Methods:

• Retrospective collection and observer-based labeling of OPGs for apex closure status.
• Evaluation of two ViT models (Base Patch16, Patch32) and three CNN models (ResNet50, VGG19, EfficientNetB0).
• Application of eight classifiers (SVM, RF, XGBoost, LR, KNN, NB, DT, MLP) with performance metric computation (accuracy, precision, recall, F1, AUC).

Main Results:

• ViT Base Patch16 384 with MLP achieved the highest accuracy (0.8462 ± 0.0330) and AUC (0.914 ± 0.032).
• ViT models demonstrated more balanced and robust performance compared to CNNs.
• EfficientNetB0 + MLP showed competitive performance but was outperformed by the best ViT model.

Conclusions:

• ViT models exhibit superior performance over CNNs for open apex classification on OPGs.
• ViT models hold promise for integration into dental radiologic decision support systems.
• Future research should explore multi-center and multimodal data for enhanced generalizability.