Tooth Segmentation and Abnormal Tooth Detection with Diagnostic Criterion in Panoramic X-Ray Images with Deep Learning Approach

Abstract

In this study, the aim is to perform tooth segmentation and abnormal tooth detection using diagnostic criteria for panoramic X-ray images. The 2023 dentex dataset was used in the research. During the research process, separate artificial intelligence models were examined for segmentation and diagnosis tasks. As a result of this study, the YOLO, U-Net, Trans-UNet, and DeepLabV3 AI models were selected, trained, and tested for the segmentation process, while customized CNN, DINOV2, ResNet, and EfficientNet models were chosen for abnormal tooth classification. At the end of the study, the YOLO model achieved the best results for segmentation with 96.45% AIoU, 94.21% AP, 95.52% AR, and 94.81% AP metrics, while the DINOV2 model obtained the best results for classification with 79.77% AA, 79% AP, 80% AR, 79% F1-score, and 64% Kappa metrics.