Respiratory motion is a major source of error in many image acquisition applications and image-guided interventions, and motion estimation techniques have been widely applied to compensate for it. Existing respiratory motion estimation methods typically reply on breathing motion models learned from certain training data. However, none of these methods can effectively handle both intra-subject and inter-subject variations of respiratory motion. In this paper, we propose a respiratory motion estimation method that directly recovers motion fields from sparsely spaced dynamic 2-D MRIs without a learned respiratory motion model. We introduce a scatter-to-volume registration algorithm to register the dynamic 2-D MRIs with a static 3-D MRI to recover dense motion fields. The proposed method was validated on 4-D MRIs acquired from 5 volunteers with breathing pattern variabilities, demonstrating significant improvements over the state of the art respiratory motion modeling method.
View on arXiv