AI-based determination of Kennedy classification and modification spaceson panoramic radiographs

Aims: The aim of this study is to automatically perform tooth segmentation, FDI (Fédération Dentaire Internationale) tooth numbering, and the Artificial Intelligence (AI)-based determination of Kennedy classification and its modifications using images obtained from panoramic dental radiographs. The study aims to help clinical decision support processes in prosthetic dentistry. Methods: The U-Net architecture was used for pixel-level tooth segmentation on panoramic dental radiographs, and different pre-trained encoder backbones were compared. Segmentation performance was evaluated using the Dice Similarity Coefficient and Intersection over Union metrics. The outputs of the best-performing model were integrated with the FDI tooth numbering system to automatically determine Kennedy classification and modification areas. Results: The results were evaluated by two clinicians specialized in prosthetic dentistry. Findings the U-Net model with the ResNet34 encoder demonstrated higher and more balanced segmentation performance compared to the other architectures. Statistical analyses revealed that the ResNet34 model was significantly superior to the other encoder configurations used in the study (p<0.05). The FDI-based analysis of the segmentation outputs showed that Kennedy classification and its modifications could be automatically determined in accordance with clinical rules. Conclusion: This study demonstrates that Kennedy classification and its modifications can be automatically determined using AI-assisted analysis of panoramic dental radiographs and provides a robust framework for decision support systems in prosthetic dentistry. Unlike prior AI-based studies that address Kennedy classification only at the basic class level, the present study explicitly and systematically incorporates modification space determination, representing a more detailed and systematic approach toward automated prosthetic assessment.