Cone-beam rotational angiography enables 3D imaging of the hepatic vasculature and is considered beneficial for guidance of transcatheter arterial chemoembolization procedures. Respiratory motion during the rotational acquisition challenges state-of-the-art reconstruction algorithms as intra-scan motion leads to inconsistencies causing substantial blurring and streaking artifacts in uncompensated reconstructions, suggesting the need for motion correction. We propose an automated method for respiratory motion estimation and compensation based on registration of an initial 3D arterial model to vesselness enhanced 2D projection images. Centerline points of the arterial tree are modeled as B-splines over time, the control point positions of which are optimized using α-expansion moves on graph cuts. This approach naturally allows for the estimation of 3D rigid translations as well as non-rigid deformations. Applied to a pre-clinical and a clinical acquisition, the proposed methods resulted in notable reductions in reprojection error and increased vessel sharpness that are reflected in less streaking and blurring artifact compared to the uncompensated case, implying superior vessel contrast. As the proposed methods are generic, future work will investigate their applicability to related rotational angiography imaging protocols, such as coronary angiography.
Keywords: Discrete optimization; Image reconstruction; Interventional imaging; Motion modeling.
Copyright © 2018 Elsevier Ltd. All rights reserved.