부산대학교 도서관

Deep learning enables fast deformable medical image registration but requires large training datasets, which are currently scarce. To overcome this, synthetic deformations can be generated to create and augment the training data. We propose a method that incorporates prior knowledge of the physiological motion to generate more realistic deformations. Specifically, our method is developed on thoracic computed tomography scans and incorporates respiratory motion. We evaluated the effect of various synthetic deformation methods on deep learning-based registration performance, achieving better performance when trained on realistic deformations, compared to when trained on random deformations. In general, the inclusion of realistic deformations, either real or synthetic, was found to be essential for achieving good registration performance.