A deep learning model using convolutional neural networks for caries detection and recognition with endoscopes.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Dental caries is a prevalent global oral health issue, particularly in regions with limited medical resources.
• Patients often delay seeking treatment until pain is severe, highlighting the need for early detection methods.
• Deep convolutional neural networks (CNNs) show promise in medical image analysis, including stomatology.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting and recognizing dental caries using endoscopic images.
• To leverage accessible endoscopic technology for improved caries monitoring.

Main Methods:

• A classification and semantic segmentation model (DeepLabv3+) was trained using 1,253 endoscopic images (194 non-caries, 1,059 caries).
• The models were developed using images from the Department of Stomatology at PLAGH.
• A 5-fold cross-validation protocol was employed for model evaluation.

Main Results:

• The classification model achieved a high Area Under the Curve (AUC) of 0.9897.
• The segmentation model demonstrated strong performance with an accuracy of 0.9843 and specificity of 0.9943.
• Key segmentation metrics included a Dice coefficient of 0.7099 and Intersection over Union (IoU) of 0.5779.

Conclusions:

• A deep learning model utilizing endoscopic images can effectively monitor dental caries.
• The developed model supports early diagnosis and timely treatment of dental caries.
• This approach offers a potential solution for accessible oral health monitoring.