Automated Image Quality Evaluation Of Panoramic Images Using Vision Transformer

To develop and test an Vision Transformer (ViT) architecture-based algorithm for detecting common technical errors in panoramic images. A total of 1,806 panoramic images were retrospectively collected and preprocessed through data normalization and data augmentation. The dataset was divided into training, validation, and test sets in a 4:1:1 ratio. Six quality-related labels were defined: foreign objects, incomplete image coverage, asymmetry, incorrect head positioning, incorrect chin positioning, and incorrect tongue positioning. A ViT-based model was trained using manual annotations as ground truth. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), and prediction speed was measured. The study received approval from the Institutional Medical Ethics Committee. The model achieved AUC values of 0.96, 0.96, 0.61, 0.62, 0.88, and 0.93 for detecting foreign objects, incomplete image coverage, asymmetry, incorrect head positioning, incorrect chin positioning, and incorrect tongue positioning, respectively. The average processing time per image was 0.03 ± 0.002 seconds, demonstrating efficient real-time performance. The proposed ViT-based model provides an effective automated solution for evaluating panoramic image quality, with high accuracy in detecting most technical errors. Its rapid processing speed and potential for real-time feedback make it a valuable tool for quality control and operator training in clinical settings. Further improvements are needed to enhance performance in detecting asymmetry and head positioning errors.