Skip to Main Content
Registration between 3-D volumes and 2-D fluoro images for Electrophysiology (EP) is a challenging task due to the lack of corresponding features between 2-D and 3-D data. This paper presents an automatic, accurate and workflow-friendly 2-D/3-D registration method specially designed for patient movement correction during EP procedures. Firstly, 2-D spines are enhanced by exploiting the temporal information in the EP fluoro sequence and by removing the moving structures (i.e. catheters) that can negatively affect the registration accuracy. Secondly, digitally reconstructed radiographs (DRRs) are computed efficiently from the segmentation of the 3-D spines using GPU-based method. Thirdly, special 2-D/3-D registration scheme is proposed, including an initial pose adjustment of the 3-D volume via automatic estimation of the spine orientation, a nonlinear histogram mapping between DRRs and X-ray images that maximizes their similarity in intensity distributions, and a gradient difference-based similarity measure that is tailored for comparing spine structures. Experiments on real clinical data demonstrate the efficacy of the proposed method.