Sketch-based Registration of 3D Cine MRI to 4D flow MRI

Abstract

Cardiac 4D Flow magnetic resonance imaging (4D Flow MRI) is a recent powerful technology that uniquely enables in-vivo acquisition of time-varying volumetric blood flow velocity field information in the three spatial dimensions over the cardiac cycle. Hence, 4D Flow MRI has emerged as an important medical diagnostic tool for evaluation of blood flow alteration in the heart chambers and great vessels. A critical requirement for accurate quantification and visualization of blood flow within the different heart chambers (e.g. the left ventricle (LV)) is the accurate anatomical context of cardiac chambers, which is missing in the 4D Flow MRI data. To tackle this problem, recent studies have proposed fusing the 4D Flow data with a complementary anatomical MRI scan (short axis 3D (multiple 2D slices) cine SSFP) through registration. However, since image registration is a non-linear optimization problem, the registration is slow and may not be accurate (e.g. the left ventricle can be incorrectly aligned to the right ventricle). To improve the registration performance and accuracy, localization techniques can be used. In this paper, we propose two sketch-based methods for effective localization of 4D Flow MRI to 3D cine MRI registration. We evaluate these methods and compare them with other localization methods.