A Validation Employing Convolutional Neural Network for the Radiographic Detection of Absence or Presence of Teeth

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Dental radiography is crucial for diagnosis and treatment planning.
• Automated object detection in medical images is an active area of research.
• Accurate tooth detection is essential for comprehensive dental assessments.

Purpose of the Study:

• To develop and evaluate an effective convolutional neural network (CNN) for detecting the presence or absence of teeth in dental panoramic images.
• To achieve high accuracy and reduce calculation times compared to manual methods.
• To create a robust system applicable across various imaging devices and patient conditions.

Main Methods:

• Utilized a dataset of 8000 dental panoramic images, manually annotated by experienced dentists.
• Employed a two-layer neural network architecture: Matterport Mask RCNN for object detection and ResNet (Atrous Convolution) for classification.
• Trained and validated the model to assess its performance in tooth detection and classification.

Main Results:

• The developed neural network achieved a high accuracy of 99.24% with a total loss of 0.76%.
• Demonstrated near-perfect accuracy in detecting teeth across images from different devices, pathologies, and age groups.
• The system proved effective in identifying the absence or presence of teeth with reduced computational time.

Conclusions:

• The proposed CNN architecture, combining Mask RCNN and ResNet, is highly effective for automated tooth detection in dental radiography.
• This AI-driven approach offers a reliable and efficient tool for clinical diagnosis, potentially improving dental care.
• The model's robustness across diverse imaging conditions and patient factors highlights its clinical applicability.