Tooth detection with deep learning method in mixed dentition

Aim: In recent years, the interest in artificial intelligence has increased due to technological advancements. The use of AI-supported applications has grown in the field of dentistry by keeping up with these technological developments. The aim of this study is to evaluate the effectiveness of a deep learning model developed using artificial intelligence in detecting and numbering teeth in panoramic radiographs taken from children in the mixed dentition period. Correct detection and numbering of teeth in children's panoramic radiographs are critical for treatment planning and orthodontic analyses. The use of deep learning algorithms aims to automate these processes with high accuracy, reducing the workload of dentists and increasing diagnostic accuracy. Materials and Methods: In this study, 1378 panoramic radiographs taken from children aged 6-12 were used. The panoramic radiographs were obtained from the radiology archive of the Department of Pediatric Dentistry, Faculty of Dentistry at Ordu University. The YOLOv8 architecture, a CNN-based model, was used for automatic tooth detection and numbering. The performance of the YOLOv8 model was evaluated using a confusion matrix. Results: The CNN-based model used for tooth detection in the mixed dentition period demonstrated high performance in detecting and numbering teeth. The precision, recall, F1-score, and mAP values calculated using the confusion matrix for each age group were as follows: For the 6-year-old group; 0.8846, 0.8949, 0.8897, 0.8914, for the 7-year-old group; 0.8973, 0.8958, 0.8966, 0.8801, for the 8-year-old group; 0.8795, 0.9313, 0.9055, 0.8773, for the 9-year-old group; 0.8734, 0.8933, 0.8832, 0.8984, for the 10-year-old group; 0.9590, 0.9721, 0.9655, 0.9672, for the 11-year-old group; 0.9460, 0.9588, 0.8966, 0.9523, and for the 12-year-old group; 0.9473, 0.9551, 0.9511, 0.9571. Conclusion:The results of our study, which aimed to detect and number deciduous and permanent teeth in panoramic radiographs taken from children in the mixed dentition period using deep learning-based artificial intelligence algorithms, are promising. The performance results of using AI in panoramic radiographs taken from pediatric patients may encourage further research, especially in the field of pediatric dentistry. Keywords: Artificial intelligence, deep learning, mixed dentition