Dental caries detection using a semi-supervised learning approach.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Early diagnosis of dental caries prevents invasive treatments.
• Dental radiography and deep learning (DL) are used for caries detection.
• Training DL models requires large annotated datasets, which are scarce in clinical settings.

Purpose of the Study:

• To develop an efficient self-training method for caries detection and segmentation.
• To leverage limited labeled data and abundant unlabeled data for improved DL model training.
• To enhance self-supervised learning strategies for dental image analysis.

Main Methods:

• A self-training approach using a teacher-student model architecture.
• Utilizing centroid cropped images and augmentation techniques for self-supervised learning.
• Training on a dataset of 141 labeled dental radiographic images.

Main Results:

• The proposed self-supervised learning strategy improved average pixel accuracy by approximately 6%.
• Mean intersection over union improved by approximately 3% compared to standard self-supervised learning.
• The method offers computational and performance gains over fully supervised and standard self-supervised methods.

Conclusions:

• The self-training method effectively utilizes limited labeled data for dental caries detection.
• The proposed strategy enhances self-supervised learning for medical image analysis.
• This approach offers a promising solution for data-scarce scenarios in dental diagnostics.