Related Experiment Video
Updated: May 25, 2025
05:49
Author Spotlight: Advancing CBCT and Digital Dental Image Integration with AI-Assisted Digitization
Published on: February 23, 2024
756
Deep Learning-Based Periapical Lesion Detection on Panoramic Radiographs
Viktor Szabó1, Kaan Orhan1,2,3, Csaba Dobó-Nagy1
1Department of Oral Diagnostics, Faculty of Dentistry, Semmelweis University, 47 Szentkiralyi Str., 1088 Budapest, Hungary.
Diagnostics (Basel, Switzerland)
|February 26, 2025
Summary
The Diagnocat AI system shows high accuracy in detecting periapical lesions (PL) on panoramic radiographs (PRs), serving as a valuable diagnostic support tool. While generally effective, its performance varies for specific teeth like canines.
Area of Science:
- Dentistry
- Radiology
- Artificial Intelligence
Background:
- Periapical lesions (PL) are common findings on panoramic radiographs (PRs).
- Accurate detection of PL is crucial for timely diagnosis and treatment planning.
- Artificial intelligence (AI) systems offer potential for improving radiographic interpretation.
Purpose of the Study:
- To evaluate the accuracy of the AI-based Diagnocat system (DC) in detecting periapical lesions (PL) on panoramic radiographs (PRs).
- To assess the performance metrics including sensitivity, specificity, and diagnostic accuracy of the AI tool.
- To investigate the influence of factors like palatoglossal air space (PGAS) and specific tooth types on AI detection accuracy.
Main Methods:
- A dataset of 616 teeth from 357 PRs was analyzed, comprising 308 teeth with and 308 without visible PL.
- Teeth were categorized into groups based on radiographic signs: caries, coronal restoration, and root canal filling.
- The Diagnocat system's convolutional neural network (CNN) performance was evaluated using sensitivity, specificity, predictive values, and diagnostic accuracy.
Main Results:
- The Diagnocat system achieved an overall sensitivity of 0.78, specificity of 1.00, and diagnostic accuracy of 0.89.
- Group 2 (teeth with coronal restoration) demonstrated the highest performance with sensitivity, specificity, and accuracy of 0.84, 1.00, and 0.95, respectively.
- The palatoglossal air space (PGAS) did not significantly affect PL detection (p=1). Lower detection accuracy was noted for central incisors, wisdom teeth, and canines.
Conclusions:
- The CNN-based Diagnocat system demonstrates significant potential as a decision-support tool for diagnosing periapical lesions on panoramic radiographs.
- The AI system can aid clinicians in radiographic assessments, improving diagnostic efficiency.
- Further refinement may be needed to enhance detection accuracy for specific tooth types, such as canines.
