Performance of a convolutional neural network algorithm for tooth detection and numbering on periapical radiographs

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate tooth detection and numbering are crucial for dental diagnostics and treatment planning.
• Manual tooth identification can be time-consuming and prone to errors.

Purpose of the Study:

• To evaluate the performance of a Faster Region-based Convolutional Neural Network (R-CNN) algorithm for automated tooth detection and numbering on periapical images.
• To assess the accuracy and efficiency of an AI-driven system in a clinical context.

Main Methods:

• A dataset of 1686 periapical radiographs was retrospectively collected.
• A pre-trained GoogLeNet Inception v3 CNN model was used for pre-processing, with transfer learning applied for training.
• The algorithm comprised jaw classification, region detection models, and a final integrated model, analyzed using a confusion matrix.

Main Results:

• An artificial intelligence algorithm (CranioCatch) was developed based on R-CNN inception architecture.
• In the test dataset of 864 teeth from 156 radiographs, 668 were correctly numbered.
• The algorithm achieved an F1 score of 0.8720, precision of 0.7812, and sensitivity of 0.9867.

Conclusions:

• The CNN algorithm demonstrated high accuracy and efficiency for automated tooth detection and numbering.
• Deep learning methods can significantly reduce clinician workload and improve dental record accuracy.
• This technology holds potential for applications in clinical dentistry and forensic science.