Detection of teeth and jaw bone from cone-beam computed tomography images using deep learning methods

Today, with the rapid development of technology, computer-based solutions in the health sector are developing rapidly. It is very important to make careful and detailed treatment planning before the treatment, especially in dental treatments. For this reason, the analysis of images obtained from cone-beam computed tomography devices is very important in dental treatments. An exemplary software has been developed in the study that automatically detects the structure of teeth and jawbone using cone-beam computed tomography images using U-Net, DeepLab V3+, and YOLO V3 deep learning architectures. The results obtained from the training of three different deep learning architectures used were evaluated according to the dice similarity coefficient performance evaluation criterion. Among the three deep learning architectures the U-Net deep learning model, 0.001 learning rate, 70 epochs numbers, and the taken batch size in each training were taken as 16 and the model was evaluated, according to the model dice similarity coefficient, with 0.9289 accuracy rate of tooth and jawbone detection.