Assessment of artificial intelligence in detecting errors on panoramic radiographs

This study aimed to evaluate the ability of artificial intelligence (AI) to detect common errors in panoramic radiographs.

This retrospective study utilized a dataset of 2,888 anonymized panoramic radiographs obtained from multiple dental imaging units. Three annotators classified the images into 3 categories: "uneven magnification," "tongue space," and "normal." Three deep learning architectures-Attention Dental X-ray, ResNet50, and MobileNet-were developed and evaluated. Model performance was assessed using accuracy, precision, sensitivity, area under the curve, and F1 score with 5-fold cross-validation. Statistical analysis was performed using Python version 3.9 (Python Software Foundation, Beaverton, USA) with the Scikit-learn, Matplotlib, Seaborn, and PyTorch libraries for model development and evaluation. McNemar's test was used for pairwise model comparisons, and Cohen's kappa was used to assess intra-rater and inter-rater reliability.

The Attention Dental X-ray model demonstrated superior performance among all evaluated models. It achieved the highest performance in classifying tongue space errors, with an average accuracy of 82.7%, whereas the ResNet50 and MobileNet models achieved accuracies of 56.9% and 64.1%, respectively.

Artificial intelligence models, particularly those incorporating attention mechanisms, show strong potential as supplementary tools for detecting errors on panoramic radiographs, improving image quality, and reducing radiation exposure. This study extends the application of AI beyond disease detection to quality assurance in dental radiography. Future research should focus on real-time integration for immediate operator feedback and on the development of automated error correction methods.