Global and Local Cardiac Functional Analysis with Cine MR Imaging: A Non-Rigid Image Registration Approach

Abstract

According to the World Health Report 2003, cardiovascular disease (CVD) made up $29.2\%$ of total global deaths, which highlights the importance of clinical cardiovascular research. Magnetic Resonance Imaging (MRI) has become an important technology to assist clinical diagnosis and treatment of cardiovascular disease that is non-invasive and radiation-free. Cine MRI can provide high-quality images of the beating heart with a good time resolution. Tagged MRI can be used to image the myocardium deformation by altering the magnetization spatially, which deforms with the myocardium during the cardiac cycle.

The quantitative evaluation of cardiac functions can be divided into three categories: volumetric analysis, geometry analysis and deformation analysis. Both volumetric and geometry analysis requires the segmentation of the myocardium in the MR images. The myocardium segmentation identifies the shape and size of the ventricles that are used to compute ventricular volumes and derive geometric parameters, such as curvature. The deformation analysis measures the myocardium strains, which reflect the contractibility and stretchability of the myocardium on a local scale.

Manual myocardium segmentation can be extremely labor-intensive due to the large number of images that need to be processed in a limited amount of time. A large number of fully automatic contouring algorithms have been proposed. But they are generally unreliable, and manual corrections are usually needed. In this dissertation, we propose a dual-contour propagation algorithm that propagates a small number of manual contours at two critical frames of the cardiac cycle to all the other time frames. Since manual contours are usually drawn at the two critical frames in most institutions, no extra labor is needed to perform the propagation. We validate our contour propagation algorithm on patient data, and it is shown to be statistically as accurate as manual contours.

Although myocardium deformation analysis is usually performed with tagged MRI, there are several disadvantages. First, the tags in tagged MR images fade quickly and can only be reliable over about half of the cardiac cycle. Second, the spatial resolution of the tag lines are limited and sparse inside the myocardium. Furthermore, tagged MR imaging is a more complicated protocol and is not as commonly available as cine MR imaging. So it will be very beneficial for clinical purposes to be able to measure myocardium strains through cine MR imaging. In this dissertation, we propose a comprehensive algorithm with several regularization schemes to measure myocardium strain for both left and right ventricles.

Both the contour propagation and myocardium strain analysis are based on non-rigid image registration. In this dissertation, we propose a comprehensive non-rigid image registration algorithm that is adaptive, topology preserving and consistent. This algorithm is computationally more efficient due to its adaptive optimization scheme. In addition, the inverse consistency and topology preservation are achieved by regularization.