| Title |
Teeth Segmentation for Orthodontics based on Deep Learning |
| Authors |
김태훈(Tae-Hoon Kim) ; 박종진(Jong-Jin Park) |
| DOI |
https://doi.org/10.5370/KIEE.2023.72.3.440 |
| Keywords |
Teeth segmentation; Deep learning; UNets; Layered UNet; Orthodontics |
| Abstract |
In this paper, we proposed a new UNet model to segment teeth from dental CBCT data for orthodontic treatment. The proposed model uses both the inter-connection and intra-connection proposed by UNet3+ and the nested convolution block proposed by UNet++ in order to utilize the skip connection structure designed to improve performance in the existing UNet series of models.
Also, in order to reduce the number of parameters to be learned, the convolution operation is used once or twice in the convolution block. For performance improvement, deep supervision was used for learning on a total of 8 nodes. The proposed layered UNet model shows better segmentation results than the existing UNet3+ and has excellent accuracy even though a small number of image data is used for learning. As a result of the simulation, the proposed layered UNet model using two convolution operations was the best with loss function values of 0.92, 0.85, and 0.79 for training, validation, and test data. |