Learning metal artifact reduction in cardiac CT images with moving pacemakers

Tanja Lossau, Hannes Nickisch, Tobias Wissel, Michael M. Morlock, Michael Grass

Abstract: Highlights•Introducting synthetic pacemaker leads into clinical data using a forward model which assures realistic insertion positions and motion trajectories.•Deep-learning-based metal shadow segmentation, sinogram inpainting and metal reinsertion without artifacts.•Purely rawdata-based, fully automatic processing pipeline for ‘Dynamic Pacemaker Artifact Reduction’ (DyPAR+).•Significant metal artifact reduction on ECG-gated as well as ungated cardiac CT data improves procedure planning for minimal invasive pacemaker lead extraction.•Superior robustness to second pass approach comprising image-based metal segmentation and inverse distance weighting for variations in contrast-enhancement, motion levels and acquisition settings.