Real time cardiac image registration during respiration: a time series prediction approach

Cardiac image registration is drawing attention for a range of merits in integrating and enhancing real-time (RT) images using a priori and complementary images of the myocardium, which might additionally be captured from other modalities. Myocardial stem cell delivery and radio-frequency ablation are some of the cases that could benefit from RT registration of high quality images. Unfortunately, most of these applications are of long duration, and must account in some manner for respiratory motion. Moreover, registration is not so keen as to compensate for these motions. Time series prediction techniques could compensate this shortcoming by proposing future approximate displacements caused by respiratory motion. In this study, we propose a three-stage framework for RT 2D into 3D cardiac image registration during respiration, composed of prior registration to extract the trend of respiratory motion and to calibrate a set of time series predictors for future motion prediction, as well RT registration to update estimated transform parameters. The proposed approach was validated in the course of four simulations and shows acceptable results for clinical circumstances.