Sector Classification of Unerupted Maxillary Canines: A Deep Learning‐Based Automated Framework Using Panoramic Radiographs

Objectives: To develop a deep learning-based framework to automate sector classification of unerupted maxillary canines (UMCs), assessing its accuracy and reliability compared to human ones. Material and Methods: One thousand five hundred twenty-eight UMCs from digital panoramic radiographs (PRs) were selected using data from the Dental Department, “San Giovanni di Dio e Ruggi d'Aragona” Hospital, Salerno. After a training session with an expert in sector classification, six dental practitioners allocated UMCs of a 20% random sample of the original set (T0) in 3 different sectors according to Kim's sector classification system. The assessment was repeated after 4 weeks (T1). The accuracy and reliability of the human in determining the position of the canines were defined based on the level of agreement between the trainer and the examiners and the intra-examiner agreement, respectively, both assessed through Cohen's K. The same radiographs were tested on different artificial intelligence (AI) models, pre-trained on the extended dataset. The best-performing model was identified based on its sensitivity and precision, and the model accuracy and repeatability were determined. Results: Regarding UMC allocation according to different sectors, the agreement between examiners and trainer was 0.78 (95% confidence interval = 0.77–0.80). The overall intra-examiner agreement was 0.85 (95% confidence interval = 0.83–0.87). DenseNet121 proved to be the best-performing model in allocating UMCs in the three different sectors, with an overall accuracy and repeatability of 76.8% and 95.3%, respectively. Conclusion: The developed framework provides an automated approach in sector classification of UMCs, whose accuracy is comparable to that of humans, but the reliability is greater.