Related Experiment Video
Updated: Jun 13, 2026
Reliability of Artificial Intelligence-Based Cone Beam Computed Tomography Integration with Digital Dental Images
Published on: February 23, 2024
Technical error identification in intraoral radiographs using multiple convolutional neural networks: an exploratory
Mariam Baghdady1,2, Jagan Kumar Baskaradoss3, Sabarinath Prasad4
1Oral and Maxillofacial Radiology, Department of Diagnostic Sciences, College of Dentistry, Kuwait University, P.O. Box: 24923, 13110, Safat, Kuwait. mariam.baghdady@ku.edu.kw.
Objectives:
High-quality intraoral radiographs are vital for accurate diagnosis and treatment planning. Operator-dependent errors can compromise image quality and reliability. Artificial intelligence (AI) offers strategies to automate the detection of suboptimal quality radiographs. This study presents the first step towards developing a comprehensive automated radiographic error detection software.
Methods:
Five convolutional neural networks (CNNs), GoogLeNet, Places-365 GoogLeNet, NASNet-Mobile, ShuffleNet, and ResNet 50, were compared for their capability to classify radiographs (n = 1600) as error-free (diagnostic), or with errors (cone cut, open-mouth and horizontal overlap). 20% of the images were used for validation, and another 20% for testing. Performance metrics, including sensitivity and specificity, were calculated to compare model performance.
Results:
The best performance metrics of the CNNs in identifying absence (normal images) or presence of technical errors (cone cut, open-mouth, and overlap errors were, Places-365 GoogLeNet (sensitivity/specificity - 0.75/0.85), GoogLeNet (sensitivity/specificity - 0.91/0.96), ResNet-50 (sensitivity/specificity - 0.86/0.97), and GoogLeNet (sensitivity/specificity - 0.83/0.82), respectively.
Conclusion:
AI-driven technical error detection in intraoral radiographs shows promise. A combination of CNNs may be the best approach to identify radiographic errors. Future research should focus on integrating multiple CNN strategies, employing explainable AI techniques, and expanding the dataset to refine real-time error detection for facilitating widespread clinical adoption.