The detection of distomolar teeth on panoramic radiographs using different artificial intelligence models

Area of Science:

• Dental Radiology
• Artificial Intelligence in Medicine
• Machine Learning Applications

Background:

• Distomolars are anomalous teeth outside the typical human dental system.
• Accurate identification of distomolars is crucial for clinical diagnosis and treatment planning.
• Panoramic radiography (PR) is a common imaging technique in dentistry.

Purpose of the Study:

• To employ convolutional neural networks (CNNs) for classifying distomolar tooth presence using panoramic radiography (PR).
• To evaluate the performance of various CNN architectures in detecting distomolars.

Main Methods:

• A dataset of 263 PRs (117 with distomolars, 146 without) was analyzed.
• Ten CNN frameworks (AlexNet, DarkNet, DenseNet, EfficientNet, GoogLeNet, ResNet, MobileNet, NasNet-Mobile, VGG, XceptionNet) were utilized.
• Transfer learning and 5-fold cross-validation were applied, with results fused for final classification.

Main Results:

• Performance was assessed using accuracy, sensitivity, specificity, and precision.
• The fusion of DarkNet, DenseNet, and ResNet achieved the highest accuracy of 96.2% in distomolar classification.
• These models demonstrated strong performance in identifying distomolar teeth.

Conclusions:

• CNNs show significant potential for accurate distomolar detection in dental radiographs.
• The fusion of ResNet, Darknet, and DenseNet architectures offers a promising approach for AI-driven dental diagnostics.
• These AI systems can assist clinicians in radiologic examinations and treatment planning.