Detection and dual-label classification of tooth number and condition in dental panoramic radiographs including deciduous teeth.

Objectives: The purpose of this study is to automatically extract the information necessary for chart recording from panoramic radiographs, to reduce the workload for dentists.

Study design: Using 1,085 dental panoramic radiographs (994 of permanent dentition and 91 of mixed dentition) taken at 10 facilities, we conducted tooth detection, numbering, and condition classification. Tooth condition was defined into five classes: natural, partial restoration, prosthetic crown, implant, and pontic. First, the YOLOv7 model was used to simultaneously detect 10 classes of deciduous teeth, 16 classes of permanent teeth, and four classes of tooth condition (excluding natural). We applied rule-based post-processing to the detected objects. Precision, Recall, and F1-score were used to evaluate our method, with an IoU (Intersection over Union) threshold set at 0.5.

Results: We achieved Precision, Recall, and F1-score of 98.51%, 98.38%, and 98.45%, respectively, in tooth numbering. In tooth condition classification, the average F1-score across the 5 classes was 95.47%.

Conclusions: Our method, which detects and classifies the tooth numbers of permanent and deciduous teeth and their tooth condition simultaneously, is expected to contribute to reducing the workload of dentists and improving accuracy.