Using Deep Learning Algorithms to Predict Dental Implant Brands from Panoramic Radiographs

Aim: The aim of this study is to predict dental implant brands from panoramic radiographs using deep learning algorithms. Material and Method: Panoramic radiographs of patients previously undergoing dental implant procedures were retrospectively screened. Radiographs were grouped into three different implant brands, with a minimum of 250 dental implants from each brand. The obtained radiographs were divided into three groups: training, validation, and test sets, with an equal distribution of implant brands in each group. 70% of the implants were used for training, 20% for validation, and 10% for the test dataset. Trained models were tested on the previously separated test set that was not used in the deep learning model training to determine the implant brand. Results: A total of 882 implants were evaluated in 220 panoramic radiographs. The study found that the accuracy of the implants tested in the deep learning model was 75% and the sensitivity was 78.26%. The accuracy of the model was 94.73%. The F1 score, which is a parameter frequently used in comparing artificial intelligence models with each other, was found to be 85.71%. Conclusion: The results of this study show that implants can be identified from panoramic radiographic images using deep learning algorithms. However, to use this system routinely in clinical practice, it is necessary to create libraries by conducting studies that include many different implant systems and a large number of images.