Caries and Restoration Detection Using Bitewing Film Based on Transfer Learning with CNNs.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Dental Diagnostics

Background:

• Dental caries is a bacterial infection requiring early detection for effective treatment.
• Current manual lesion detection on radiographic examinations is time-consuming and relies heavily on professional experience.
• Advancements in AI imaging offer potential to enhance accuracy and efficiency in dental diagnostics.

Purpose of the Study:

• To develop and evaluate a caries and lesions area analysis model using convolutional neural networks (CNNs).
• To improve the accuracy and speed of identifying caries and restorations in dental bitewing images.
• To provide objective data for automatic diagnosis and treatment planning, supporting precision medicine.

Main Methods:

• Utilized Gaussian high-pass filter and Otsu's threshold for image enhancement.
• Developed a novel method for single-tooth extraction from bitewing images, including preprocessing, cropping, and masking.
• Trained and compared four common neural networks (AlexNet, GoogleNet, Vgg19, ResNet50) for caries and restoration identification.

Main Results:

• The proposed AlexNet model achieved high accuracy: 95.56% for restoration judgment and 90.30% for caries judgment.
• The image processing steps successfully isolated individual teeth, enhancing the analysis model's performance.
• The AI model demonstrated effectiveness in identifying caries and restorations from bitewing radiographs.

Conclusions:

• The developed CNN-based model shows significant potential for automatic diagnosis of dental caries and restorations.
• This AI approach can assist dentists in making more accurate judgments and optimizing treatment planning.
• The study paves the way for developing automated bitewing film analysis tools for improved dental care.