Detection of cephalometric points with artificial intelligence algorithms improved by deep learning method on posteroanterior cephalometric (Frontal) images obtained from cone-beam computed tomography images

Aim: The aim of this study is to investigate the success of cephalometric point detection using a special artificial intelligence algorithm on posteroanterior cephalometric images obtained from cone beam computed tomography images. Material and Method: The data set of our study is to create skeletal and dental cephalometric points marked using a special artificial intelligence algorithm on posteroanterior cephalometric radiographs obtained from cone beam computed tomography images taken before orthodontic treatment of 1295 patients. Labeling of cephalometric points on the images was done using CranioCatch labeling software (CranioCatch, Eskisehir, Turkey). The training of the artificial intelligence model was carried out using 300 Epochs with the CNN-based deep learning method applied to PyTorch. The learning rate of the model was determined as 0.0001. Results: In the artificial intelligence model trained in the study, the highest SDR value was found at Antegonial Notch, L, Antegonial Notch, R and Gonion, L points. The lowest SDR value was observed at Crista Galli. The SDR value showed a success rate of over %80, except for 6 points in the 2 mm interval. Conclusion: Our study is very important for the development of deep learning-based CBCT reporting systems to be made in the future. It is thought that these systems will play a role as a decision support mechanism by saving time for physicians in their clinical routine. At the same time, it is estimated that during the determination of cephalometric points, it will be very helpful in minimizing the differences between observers and the inconsistencies that may occur in the evaluations of the observers at different times.