Automated permanent tooth detection and numbering on panoramic radiograph using a deep learning approach

Objective: This study aimed to assess the performance of the deep learning (DL) model for automated tooth numbering in panoramic radiographs. Study Design: The dataset of 500 panoramic images was selected according to the inclusion criteria and divided into training and testing data with a ratio of 80%:20%. Annotation on the data set was categorized into 32 classes based on the dental nomenclature of the universal numbering system using the LabelImg software. The training and testing process was carried out using You Only Look Once (YOLO) v4, a deep convolution neural network model for multiobject detection. The performance of YOLO v4 was evaluated using a confusion matrix. Furthermore, the detection time of YOLO v4 was compared with a certified radiologist using the Mann-Whitney test. Results: The accuracy, precision, recall, and F1 scores of YOLO v4 for tooth detection and numbering in the panoramic radiograph were 88.5%, 87.70%, 100%, and 93.44%, respectively. The mean numbering time using YOLO v4 was 20.58 ± 0.29 ms, significantly faster than humans (P < .0001). Conclusions: The DL approach using the YOLO v4 model can be used to assist dentists in daily practice by performing accurate and fast automated tooth detection and numbering on panoramic radiographs.