Automated Tooth Detection and Caries Identification in CBCT With Deep Learning.

Area of Science:

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

Background:

• Automated localization and numbering of carious teeth in CBCT images is underexplored.
• Deep learning shows potential for caries diagnosis but requires robust tooth localization.

Purpose of the Study:

• To develop a two-stage deep learning framework for tooth detection, numbering, and caries identification in CBCT images.
• To provide a technical basis for automated analysis supporting opportunistic caries screening.

Main Methods:

• Retrospective study using CBCT images from 65 patients.
• YOLOv3 and Cascade R-CNN were compared for tooth detection.
• DenseNet169, MobileNet_V2, and ResNet50 were evaluated for caries identification.

Main Results:

• YOLOv3 demonstrated superior tooth detection performance (P < .0001).
• DenseNet169 achieved the best caries identification with balanced accuracy of 0.7414 and MCC of 0.6074.
• The integrated framework showed acceptable overall performance.

Conclusions:

• The proposed two-stage framework shows promising performance for detecting, numbering, and identifying caries in CBCT images.
• This supports the feasibility of opportunistic screening using CBCT scans for non-caries indications.
• Clinical utility requires further multi-center and prospective validation.