The U-Net Approaches to Evaluation of Dental Bite-Wing Radiographs: An Artificial Intelligence Study

1. Introduction

• Overview: The study investigates the use of artificial intelligence (AI), particularly the U-Net convolutional neural network (CNN), in the evaluation of dental bite-wing radiographs.
• Significance: Bite-wing radiographs are crucial in dental diagnostics. Utilizing AI can enhance efficiency and accuracy in detecting dental conditions.
• AI in Radiology: AI, especially deep learning methods, has shown promise in medical diagnostics, reducing error rates compared to human observers.

2. Materials and Methods

• Data Collection: 500 bite-wing radiographs from Eskisehir Osmangazi University's Faculty of Dentistry were used. High-quality images were selected, excluding those with errors.
• Imaging: Radiographs were taken using the ProX periapical X-ray unit and stored in the CranioCatch labeling software.
• Image Labeling: Five dental diagnoses were labeled: dental caries, dental crown, dental pulp, dental restorative filling material, and dental root canal filling material.
• Model Development: The U-Net model was implemented using the PyTorch library. The model was trained on a dataset split into 80% training, 10% testing, and 10% validation.
• Training and Testing: The model was trained for 200 epochs with a learning rate of 0.0001. A computer with 16 GB RAM and an NVIDIA GeForce GTX 1660 TI graphics card was used.

3. Results

• Performance Metrics: The model's performance was evaluated using sensitivity, precision, and F1 scores. Results for each dental condition were promising:
  • Dental Caries: F1 score of 0.8818, sensitivity of 0.8235, precision of 0.9491.
  • Dental Crown: F1 score of 0.9629, sensitivity of 0.9285, precision of 1.
  • Dental Pulp: F1 score of 0.9631, sensitivity of 0.9843, precision of 0.9429.
  • Restorative Material: F1 score of 0.9714, sensitivity of 0.9622, precision of 0.9807.
  • Root Canal Filling: F1 score of 0.9722, sensitivity of 0.9459, precision of 1.

4. Discussion

• Significance of Results: The AI model shows high accuracy and reliability in evaluating bite-wing radiographs. This can significantly aid dentists in clinical settings by reducing their workload and improving diagnostic efficiency.
• Future Implications: The study suggests further refinement and validation of AI models with larger datasets and different types of radiographs.

5. Conclusion

• Summary: The U-Net based AI model demonstrates the potential to automatically evaluate dental bite-wing radiographs accurately.
• Clinical Impact: Such AI models can support dental professionals by providing quick and reliable diagnostic assistance, enhancing patient care.