Keywords: Tooth symptoms, CBCT, Dental prior, Symptom shape loss
TL;DR: TSNet improves CBCT image segmentation with Dental Prior Guiding Data Augmentation (DPGDA) and Dental Symptom Shape Loss (DSSL), outperforming state-of-the-art methods across diverse symptom datasets.
Abstract: Automated dental diagnosis requires accurate segmentation of tooth from cone-beam computed tomography (CBCT) images. However, existing segmentation methods often overlook incorporating prior information and symptoms of teeth, which can cause unsatisfactory segmentation performance on teeth with symptoms. To this respect, we propose Tooth Symptom Network (TSNet), consisting of Dental Prior Guiding Data Augmentation (DPGDA) and Dental Symptom Shape Loss (DSSL), to improve segmentation performance for teeth with different clinical symptoms. Experiments show that TSNet outperforms all state-of-the-art methods across datasets with all kinds of symptoms with an average increase of 1.13\% in Dice and 2.00\% in IoU.