Tooth Instance Segmentation on Panoramic Dental Radiographs Using U-Nets and Morphological Processing

Automatic teeth segmentation in panoramic x-ray images is an important research subject of the image analysis dentistry. In this study, we propose a post-processing stage to obtain map which objects are separated, and apply technique tooth instance with U-Net network. The consists grayscale morphological filtering operations, applied sigmoid output network before binarization. A dice overlap score 95.4 - 0.3% obtained overall segmentation. proposed stages reduce mean error count 6.15%, whereas without 26.81%. performances both counting highest literature, our knowledge. Moreover, achieved by using relatively small training dataset, 105 images. Although aim study segment instances, presented method applicable similar problems other domains, such as separating cell instances