Classification of Dental Radiographs Using Deep Learning.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Deep Learning Architectures

Background:

• Dental radiography is crucial for diagnosis.
• Automated classification of radiographic images can improve efficiency.
• Evaluating deep learning models for dental image analysis is an active research area.

Purpose of the Study:

• To compare the performance of three deep learning architectures for classifying dental radiographs.
• To assess the accuracy of ResNet, baseline CNN, and capsule networks in distinguishing between panoramic, bitewing, periapical, and cephalometric images.

Main Methods:

• A large dataset of 31,288 panoramic, 43,598 periapical, 14,326 bitewing, and 1176 cephalometric radiographs was used.
• Active learning with ResNet-34 was employed for iterative annotation of a subset of images.
• Performance was evaluated using stratified k-fold cross-validation, with Gradient-weighted Class Activation Mapping (Grad-CAM) for visualizations.

Main Results:

• All models achieved high accuracy (>98%).
• ResNet demonstrated significantly superior accuracy, F1-score, precision, and sensitivity compared to baseline CNN and CapsNet (p < 0.05).
• Misclassifications were most frequent between bitewing and periapical radiographs, with distinct activation patterns observed for each type.

Conclusions:

• Deep learning models, particularly ResNet, show high potential for accurate dental radiograph classification.
• The image features identified by the models align with expert clinical reasoning.
• AI-powered tools can assist in the efficient and accurate analysis of dental radiographic data.