Related Experiment Video
Updated: May 14, 2026
08:02
Novel Quantification Protocol for Cardiovascular Calcification Progression Using Longitudinal MicroPET/MicroCT Images
Published on: November 15, 2024
Transfer learning models in the detection of pulp calcifications- A preliminary study
Shishir Shetty1, Wael Talaat1, Sausan AlKawas1
1Department of Oral and Craniofacial Health Sciences, College of Dental Medicine, University of Sharjah, Sharjah, United Arab Emirates.
Journal of Oral Biology and Craniofacial Research
|May 13, 2026
Summary
Transfer learning models effectively detect pulp calcifications (PCs) in dental radiographs. The VGG16 model demonstrated superior performance in identifying PCs on cropped panoramic radiographs (PRs), offering a promising tool for endodontic treatment planning.
Area of Science:
- Dentistry
- Radiology
- Artificial Intelligence
Background:
- Pulp calcifications (PCs) significantly impact endodontic treatment outcomes.
- Radiographic modalities are essential for detecting PCs.
- Transfer learning models show potential for analyzing dental radiographs with limited computational resources.
Purpose of the Study:
- To evaluate the efficacy of transfer learning models in detecting pulp calcifications (PCs).
- To assess model performance on cropped panoramic radiographs (PRs).
Main Methods:
- 240 cropped panoramic radiographs (PRs) (120 with PCs, 120 without) were analyzed.
- Images underwent Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing and augmentation.
- Pre-trained models (VGG16, ResNet101V2, MobileNetV2) were fine-tuned for classification.
Main Results:
- VGG16 achieved the highest test accuracy (0.85) and AUC (0.93).
- VGG16 demonstrated strong precision (0.84), recall (0.87), and F1-score (0.86).
- ResNet101V2 and MobileNetV2 showed lower test accuracies (69% and 50%, respectively).
Conclusions:
- The VGG16 transfer learning model is highly effective for detecting PCs in cropped PRs.
- Generalizability is limited due to cropped images; future research will focus on uncropped PRs and larger datasets.
