In permanent dentition determination the ideal extraction time of first molar teeth by evaluation of root formation levels of mandibular second molars with arti̇ficial intelligence

Permanent first molars, which are the most carious teeth and are usually lost at an early age due to their poor prognosis. Determining the appropriate time for extraction of first molars is important for the physical development of children. In the Demirjian method, which classifies the calcification stages of teeth, stage E is the ideal time for the decision to extract permanent first molars and corresponds to the age range of 8-10 years in children. In our thesis study, it was aimed to automatically classify the calcification stages of Demirjian method using convolutional neural networks with a model based on artificial intelligence and deep learning. In this study, teeth numbered 37 and 47 were labeled with segmentation and detection methods on 1106 panoramic radiographs. CranioCatch labeling software (CranioCatch, Eskişehir, Turkey) was used to label the teeth on the images. Eight datasets were created for the 8 stages of the Demirjian method from A to H. It was trained using the YOLOv5 model. The detection method yielded results with approximately 0.990 sensitivity, 0.917 precision and 0.952 F1 score, while the segmentation method yielded results with 0.985 sensitivity, 0.921 precision and 0.952 F1 score. According to the results of our study, the developed deep learning models were found to be successful in detecting/predicting the stages of the Demirjian method, which we used to determine the ideal extraction time of permanent first molars.