Show abstract - PDF - Reviews

Panoramic radiographs, also known as orthopantomograms, are routinely used by dentists to get a better understanding of the patient\'s oral condition and perform orthodontics; however, most dentists\' use of them is rough which may cause medical negligence due to physician burnout and time limitation. To reduce the dentists\' workflow, we define a mission to segment five orals structures including Alveolarcrest, Condyle, Neuraltube, Sinusmaxillaris, and Teeth on Panoramic radiographs. A Cascaded Multi-scale Mask2former(CMMask2former) method is proposed for this task. For the small objects, we specially design multi-scale masked attention for the mask area. Overall, the whole structure is designed in a cascade way of two stages for localization and prediction. Our results demonstrate a superior prediction performance compared to other methods.
Hide abstract