Artificial intelligence for caries detection: a novel diagnostic tool using deep learning algorithms.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Deep Learning for Diagnostics

Background:

• Dental caries detection from periapical radiographs is crucial for timely intervention.
• Current diagnostic methods can be subjective and time-consuming.
• Advancements in deep learning offer potential for automated analysis.

Purpose of the Study:

• To develop and evaluate an automated dental caries detection tool.
• To utilize a novel convolutional neural network (CNN) architecture, ResNet+SAM.
• To compare the model's performance against traditional CNNs and human dentists.

Main Methods:

• A dataset of 4278 annotated periapical radiographs was used to train the ResNet+SAM model.
• Performance was evaluated using metrics like F1 score, accuracy, and Area Under the Curve (AUC).
• Gradient-weighted Class Activation Mapping (Grad-CAM) visualized model attention.

Main Results:

• ResNet+SAM achieved high performance with an average F1 score of 0.886 and accuracy of 0.885.
• The model outperformed junior dentists in accuracy.
• Assisted dentists showed improved metrics and enhanced interobserver agreement.

Conclusions:

• The ResNet+SAM model demonstrates significant potential for accurate dental caries identification.
• This AI tool can serve as valuable clinical decision support, reducing dentist workload.
• Automated detection can enhance diagnostic consistency and efficiency in dental radiology.