An Explainable Deep Learning Model to Prediction Dental Caries Using Panoramic Radiograph Images

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Deep Learning for Healthcare

Background:

• Dental caries is a prevalent global oral health issue impacting quality of life.
• Current machine learning models for caries detection lack explainability, limiting clinical adoption.
• Explainable AI is crucial for physician trust and effective integration into dental practice.

Purpose of the Study:

• To develop and evaluate an explainable deep learning model for accurate dental caries detection.
• To compare the performance of prominent pre-trained models (EfficientNet-B0, DenseNet-121, ResNet-50) for this task.
• To provide visual explainability through heat maps for improved diagnostic confidence.

Main Methods:

• Utilized three pre-trained deep learning models: EfficientNet-B0, DenseNet-121, and ResNet-50.
• Input panoramic dental images to classify caries and generate explanatory heat maps.
• Evaluated model performance on 562 subjects' panoramic images using accuracy, sensitivity, and F1-score.

Main Results:

• All tested models demonstrated high performance in detecting dental caries.
• ResNet-50 slightly outperformed EfficientNet-B0 and DenseNet-121.
• ResNet-50 achieved 92.00% accuracy, 87.33% sensitivity, and 91.61% F1-score.
• Generated heat maps accurately localized suspected caries regions.

Conclusions:

• The proposed explainable deep learning model accurately and reliably diagnoses dental caries.
• Heat maps enhance model transparency, allowing dentists to validate findings.
• This approach can assist dentists in reducing misclassification and improving patient care.