Comparative Analysis of Pixel-Based Segmentation Models for Accurate Detection of Impacted Teeth on Panoramic Radiographs

Accurate detection of impacted teeth in panoramic radiographs is critical for effective diagnosis and treatment planning in dentistry. Traditional segmentation methods often face challenges in achieving accurate detection due to the anatomical complexity and variability of dental structures. This study aims to address these limitations by performing a comprehensive comparative analysis of four advanced pixel-based segmentation models - U-Net, FPN, PSPNet and LinkNet - integrated with ten different backbone architectures. Using a meticulously annotated dataset of 407 high-resolution panoramic radiographs, the models were rigorously trained and evaluated using robust performance metrics, including accuracy, precision, recall, F1 score, and Intersection over Union (IoU). Among the configurations tested, the U-Net model with an EfficientNetB7 backbone achieved the highest performance, with an average IoU score of 85.29%, demonstrating superior accuracy and reliability. The main contributions of this study are the in-depth comparison of state-of-the-art segmentation models, the identification of the most effective architectures tailored for dental radiograph segmentation, and new insights into the advantages of pixel-based approaches over region-based methods commonly used in previous studies. These findings highlight the strengths and limitations of each model, providing practical guidance for researchers and clinicians in selecting appropriate solutions for impacted teeth detection. In addition, the study highlights the potential for future advances through hybrid approaches and customized model designs to further improve detection accuracy and clinical applicability. As a result, this research demonstrates the transformative potential of integrating artificial intelligence into dental diagnostics, paving the way for more accurate, efficient and scalable solutions to improve clinical decision-making.