Related Experiment Video
Updated: Aug 7, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Deep Learning-Based Detection of Simulated Root Resorption in Scenarios Involving Image-Degrading Artifacts: An in
Orlando Aguirre Guedes1, Letícia Junqueira de Pádua Sesti Gomes Moussa1, Lucas Rodrigues de Araújo Estrela1
1Department of Oral Sciences, School of Dentistry, Evangelical University of Goiás, Anápolis, Goiás, Brazil.
Summary
A deep learning model accurately detects external root resorption (ERR) in dental X-rays and CBCT scans. Performance varied by imaging method, with periapical radiography achieving perfect accuracy.
Area of Science:
- Dentistry
- Medical Imaging
- Artificial Intelligence
Background:
- External root resorption (ERR) is a common endodontic complication.
- Accurate detection of ERR is crucial for treatment planning and prognosis.
- Existing imaging modalities can be limited by artifacts and resolution.
Purpose of the Study:
- To develop and evaluate a deep learning-based convolutional neural network (CNN) model for detecting external root resorption (ERR).
- To assess the model's performance across different imaging modalities, including periapical radiography and cone-beam computed tomography (CBCT), with varying image processing techniques.
- To investigate the impact of image-degrading artifacts on the CNN's diagnostic accuracy for ERR detection.
Main Methods:
- A deep learning model (AlexNet CNN) was adapted using transfer learning for four-class image classification.
- 480 bovine incisors were prepared with varying conditions of root canal treatment and ERR.
- Images were acquired using periapical radiography and CBCT, with CBCT images processed with and without the Blooming Artifact Reduction (BAR 1) algorithm and a 3D reconstruction tool.
Main Results:
- The CNN model achieved perfect classification accuracy (100%) for external root resorption detection using periapical radiography.
- CBCT imaging demonstrated high accuracy, with 95.83% without BAR 1 and 97.92% with BAR 1.
- CBCT with 3D reconstruction showed reduced performance (73.96%), especially in endodontically treated teeth without resorption.
Conclusions:
- Deep learning-based CNN models show high diagnostic performance for detecting external root resorption.
- Image acquisition and processing protocols significantly influence the diagnostic accuracy of AI models for ERR detection.
- Periapical radiography and optimized CBCT protocols show promise for AI-assisted ERR detection.