Classification of Approximal Caries in Bitewing Radiographs Using Convolutional Neural Networks

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Dental caries, particularly approximal caries, pose diagnostic challenges due to their location.
• Bitewing radiography is crucial for identifying approximal caries, but interpretation errors can occur.
• Computational methods offer potential solutions for improving caries diagnosis.

Purpose of the Study:

• To develop and evaluate a novel method combining image processing and CNNs for identifying and classifying approximal dental caries.
• To assess the performance of Inception and ResNet architectures in detecting caries severity from bitewing radiographs.

Main Methods:

• Acquired 112 bitewing radiographs and extracted individual tooth images.
• Applied data augmentation and trained CNN classification models (Inception, ResNet) on expert-labeled images.
• Evaluated models using learning rates of 0.1, 0.01, and 0.001 over 2000 iterations.

Main Results:

• The Inception model with a 0.001 learning rate achieved the highest accuracy of 73.3% on the test set.
• The proposed method demonstrated promising results in identifying approximal dental caries and their severity.

Conclusions:

• The developed AI-powered method shows potential as a tool to assist dentists in evaluating bitewing radiographs for approximal caries.
• This approach could aid in defining lesion severity and guiding appropriate treatment decisions.