3D Fourier Domain Adaptation for Improving CBCT Tooth Segmentation Under Scanner Parameter Shift.

Convolutional neural network based segmentation models have shown success in teeth segmentation from cone beam computed tomography (CBCT) scans. However, trained models often fail to generalize to new acquisitions when scanner protocols shift and upgrade. This problem is well-known by the machine learning community as domain shift. To address this problem in an unsupervised manner, we demonstrate the first time application of 3D Fourier Domain Adaptation of a tooth segmentation model in a source domain for an adapted target domain. Our experiments demonstrate that the proposed domain adaptation method can significantly improve the segmentation performance for the target domain.