A Deep Learning Model with Attention Mechanism for Dental Image Segmentation

Radiological imaging is a frequently used procedure in dental treatments. It provides information to the physician about areas of the tooth that cannot be seen from the outside. Digital radiological images can be processed with advanced computer vision techniques. In recent years, deep learning models with attention mechanisms which are mainly developed for natural language processing, have been applied to computer vision studies. In this study, three deep learning models, Vision Transformer (ViT), Segmenter and ConvNeXt were used on the segmentation of teeth and maxillomandibular region. The performance results were better than the U-Net and other benchmark models that are widely used in medical image segmentation. The IoU performance of the models, ConvNeXt, Segmenter and ViT, for the teeth segmentation was 90.77, 91.86, 92.63 respectively. In the maxillomandibular region segmentation IoU results of the models were 92.0, 95.56, 77.51.