Recognizing Occlusal Caries in Dental Intraoral Images Using Deep Learning.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Dental caries detection is crucial for timely treatment.
• Accurate classification across the International Caries Detection and Assessment System (ICDAS) scale is challenging.
• Intra-oral imaging offers a direct view of tooth surfaces.

Purpose of the Study:

• To evaluate a Deep Learning model (Mask R-CNN) for detecting and classifying dental caries.
• To assess performance across the full 7-class ICDAS scale using in-vivo dental images.
• To explore the utility of superpixels segmentation in annotation and evaluation.

Main Methods:

• Utilized a dataset of 88 in-vivo dental images from an intra-oral camera.
• Applied Mask R-CNN, a Deep Learning model, without image pre-processing.
• Employed superpixels segmentation for expert annotations and classifier evaluation.
• Incorporated transfer learning and data augmentation during model training.

Main Results:

• The Mask R-CNN model demonstrated capability in detecting and classifying dental caries.
• Performance was assessed across the entire 7-class ICDAS scale.
• The study achieved results without requiring image pre-processing steps.

Conclusions:

• Deep Learning, specifically Mask R-CNN, shows promise for automated dental caries detection and classification.
• Superpixels segmentation can be effectively used for annotation and evaluation in this context.
• Further improvements are possible through classifier fine-tuning and larger datasets.