A novel deep learning-based pipeline architecture for pulp stone detection on panoramic radiographs

Area of Science:

• Dentistry
• Radiology
• Artificial Intelligence

Background:

• Pulp stones are ectopic calcifications within dental pulp tissue.
• Accurate detection of pulp stones is crucial for diagnosis and treatment planning.
• Current detection methods can be labor-intensive and prone to errors.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based two-stage pipeline architecture for detecting pulp stones on panoramic radiography images.
• To address the challenge of limited annotated data in medical image analysis.

Main Methods:

• A two-stage deep learning approach was implemented, utilizing YOLOv8 for tooth localization and ResNeXt for pulp stone classification.
• The study analyzed 375 panoramic radiography images.
• Performance was rigorously assessed using metrics such as accuracy, precision, recall, F1 score, false-negative rate, and false-positive rate.

Main Results:

• The proposed deep learning pipeline achieved high performance metrics, including 95.4% accuracy, 97.1% precision, and a 96.6% F1 score.
• The system demonstrated superior performance compared to existing pulp stone detection methods.
• The model performed effectively despite a limited amount of annotated training data.

Conclusions:

• This study presents the first pipeline architecture for pulp stone detection on panoramic images using deep learning.
• The developed system offers a practical solution for automatic detection of pulpal calcifications, especially beneficial for dental students and new practitioners.
• The approach validates the feasibility of using deep learning with limited data for dental radiographic analysis.