Artificial intelligence system for classification of temporomandibular joint osteoarthritis on CONE-BEAM CT images

Aim: The purpose of this study is to develop a method that will save time and convenience for physicians in the diagnosis of the disease by using an artificial intelligence model to classify temporomandibular joint osteoarthritis and segmentation of the temporomandibular joint, which affects a significant majority of the population, on CBCT images. Material and Method: This study assessed the performance of the YOLOv5 architecture artificial intelligence model in the segmentation of the TMJ and the classification of osteoarthritis on 2000 sagittal slices (Images containing 500 healthy, 500 erosion, 500 osteophytes, 500 flattening) obtained from CBCT DICOM images of 290 patients. The temporomandibular joint was identified on the images using the labeling program CranioCatch (CranioCatch, Eskişehir, Turkey). The complexity matrix method was used to determine the model's success metric. Results: Sensitivity, precision and F1 score of the YOLOv5 model for temporomandibular joint segmentation, respectively; 1, 0.9953, 0.9976. The model's AUC value for temporomandibular joint segmentation is 0.9723. The accuracy value for the temporomandibular joint segmentation of the model was found to be 0.9953. For the classification of temporomandibular joint osteoarthritis, the sensitivity, precision and F1 scores of the model were; 1, 0.7678, 0.8686. The accuracy value found for classification is 0.7678. Conclusion: It is thought that the study can be a support mechanism that will save physicians time and convenience in the clinical and radiology routine in the diagnosis of temporomandibular joint osteoarthritis. Our work can be further developed with future studies. Key words: Deep learning, cone beam computed tomography, temporomandibular joint, temporomandibular joint osteoarthritis, artificial intelligence.