Deep Learning for Caries Detection and Classification.

Area of Science:

• Artificial Intelligence in Dentistry
• Radiographic Interpretation
• Deep Learning for Medical Imaging

Background:

• Deep learning (DL) demonstrates significant diagnostic capabilities in radiology.
• Accurate detection and classification of dental caries are crucial for effective treatment.
• Current diagnostic methods rely heavily on expert interpretation of radiographic images.

Purpose of the Study:

• To employ DL methods for detecting caries lesions on dental panoramic films.
• To classify caries lesions based on their radiographic extension (dentin depth).
• To compare the DL model's performance against expert dentists.

Main Methods:

• 1160 dental panoramic films were analyzed, with caries lesions marked by dentists to create a reference dataset.
• A convolutional neural network (nnU-Net) was used for caries lesion detection.
• DenseNet121 was utilized for classifying lesion depth (D1, D2, D3), with performance metrics compared to six expert dentists.

Main Results:

• nnU-Net achieved high performance in caries lesion segmentation (IoU: 0.785, Dice: 0.663) and detection (accuracy: 0.986, recall: 0.821).
• DenseNet121 demonstrated strong accuracy in classifying caries depth (D1: 0.957, D2: 0.832, D3: 0.863).
• The DL models' performance metrics (accuracy, precision, recall, NPV, F1-score) were statistically similar to those of experienced dentists.

Conclusions:

• Deep learning methods exhibit comparable performance to expert dentists in detecting and classifying dental caries on panoramic radiographs.
• The study highlights the potential of AI in enhancing dental disease diagnosis.
• Further research is recommended to explore the clinical impact of these DL models on diagnostic and treatment decisions.