Authors
Cansu Görürgöz, Kaan Orhan, Ibrahim Sevki Bayrakdar, Özer Çelik, Elif Bilgir, Alper Odabaş, Ahmet Faruk Aslan, Rohan Jagtap
Journal
Dentomaxillofacial Radiology, 2022
Objectives
The study aimed to evaluate the performance of a Faster Region-based Convolutional Neural Network (R-CNN) algorithm for the detection and numbering of teeth on periapical radiographs.
Methods
- Data Collection: The study retrospectively collected 1686 periapical radiographs from patients.
- Model Training: A pre-trained GoogLeNet Inception v3 CNN model was used for pre-processing and training the dataset using transfer learning techniques.
- Algorithm Components:
- Jaw Classification Model
- Region Detection Models
- Final Algorithm integrating all models
- Performance Metrics: Sensitivity, precision, true-positive rate, and false-positive/negative rates were calculated using a confusion matrix.
Results
- The AI algorithm, named CranioCatch, was designed based on the R-CNN inception architecture to detect and number teeth automatically on periapical images.
- Out of 864 teeth in 156 periapical radiographs, 668 teeth were correctly numbered.
- Performance Metrics:
- F1 Score: 0.8720
- Precision: 0.7812
- Sensitivity: 0.9867
Conclusion
The study demonstrated the CNN algorithm's high accuracy and efficiency in detecting and numbering teeth. Deep learning-based methods can help clinicians by reducing workload, improving dental records, and decreasing turnaround time for urgent cases. Additionally, this technology could contribute significantly to forensic science.
