Background and Aim: Artificial intelligence is machine systems that imitate human intelligence and can improve itself. In the present study, it was aimed to evaluate the success of deep learning algorithms, which is an artificial intelligence system, in the determination of periodontal status on dental bitewing radiographs.
Methods: In the current study, a total of 434 dental bitewing radiograph images taken from patients with periodontitis were used. Total alveolar bone losses, horizontal bone losses, vertical bone losses, furcation defects and dental calculus around maxillary and mandibular teeth on all radiographs were labeled with the segmentation method. All images were sized and split into datasets. The development of artificial intelligence models was carried out with the U-Net architecture and the confusion matrix system was used for all statistical evaluations.
Results: In all radiographs, a total of 859 total alveolar bone loss, 2215 horizontal bone loss, 340 vertical bone loss, 108 furcation defects and 508 dental calculus labeling were performed. When the success of the developed models was evaluated (50% Intersection over Union value), the sensitivity, precision and F1 score results for total alveolar bone loss were found as 1, 0.94 and 0.96, respectively. In determining the type of bone destruction for each interdental space, it was observed that the sensitivity, precision and F1 score results for horizontal bone destruction were 1, 0.92 and 0.95, respectively, but the system could not determine the vertical bone destructions. Sensitivity, precision, and F1 score results for dental calculus were 1.0, 0.7, 0.82, and for furcation defect were 0.62, 0.71, and 0.66 (respectively).
Conclusions: It is seen that artificial intelligence systems are quite successful in radiographic image interpretation. The results of the present study show that these systems can be used for periodontal status determination in dental radiographs in the future.