Ischemic heart disease is a major clinical problem. Myocardial injury from ischemic heart disease is often regional, and it is the fundamental goal of many forms of cardiac imaging and image analysis methods to measure the regional function of the left ventricle (LV), through LV wall motion, thickening, and strain measurement, in an effort to isolate the location, severity and extent of ischemic or infarcted myocardium.
However, many existing methods[42][35][21][13][10][4] have been hampered in making accurate and reliable quantitative regional LV function measurements over the entire cardiac cycle due to the fact that the heart is a non-rigid moving object that rotates, translates, and deforms in 3D space, whereas most standard approaches rely on 2D image sequence data(projections of 3D object onto the fixed 2D planes, or 2D sections). Furthermore, most of these approaches(e.g. [42][21][10]) often only use the end-diastolic(ED) and end-systolic(ES) image frames, while the LV actually goes through a temporal wave of contraction and expansion, and the asynchrony of surface motion or LV thickening from time to time and region to region may be indicative of ischemia. Unfortunately, even several attempts[13][4] that do aim at motion measurement over the entire cardiac cycle are also plagued by the need to set up a reference system that measures motion as if it
Work in the general area of 3D quantitative analysis of cardiac motion has been relatively minimal[31][23][18], though several other efforts in the general vision community are of related interest. Song and Leahy[36] have adapted the dense-field optical flow approach to include fluid flow models and have operated on 3D data. Goldgof et al[24] have been pursuing shape matching ideas similar to ours, though they primarily use Gaussian curvature under conformal stretching models. Pentland[29] and Terzopoulos[39] have been investigating non-rigid motion models, using modal finite element analysis and deformable superquadrics respectively, that might be useful for cardiac analysis also. And in [27], Ayache et al unify these two approaches([39][29]) to segment and track the object simultaneously. But all three approaches ([27][39][29]) assume that a set of corresponding landmarks are known before-han
Other alternative strategies to attempt to follow point-wise LV motion have been to artificially create distinct features that are visible in image data. Implanted myocardial markers[35][11] are accepted as a gold standard by which myocardial motion can be assessed, and they will be used as part of validation process in our study. However, this invasive technique cannot be used clinically on the study of human heart. A recent non-invasive alternative has been the MR tagging of myocardium[43][5], which creates a magnetization grid that tags the underlying tissue, and uses the grid deformation to follow the tissue movement over a gated sequence[34]. But the magnetic tags tend to decay over time, which limits the approach's ability to track motion over the entire cardiac cycle. Another new approach for point tracking is the use of phase contrast MR images to decipher local velocity which in turn can be integrated to estimate trajectories of individual points over timeIt is our hypothesis that we can use the shape of the endocardial and epicardial surfaces to track the 4D trajectories of a dense field of points, which sample the surfaces at any given time instant, over the entire cardiac cycle, based on locating and matching differential geometric landmark features and using a mathematical optimization reasoning strategy to combine local coherent smoothness model with data-derived information[1][17]. We can derive more accurate and reliable LV surface motion and wall thickening measures from these trajectories, which would be of substantial clinical value for studying the location, severity and extent of ischemic heart disease. Our method would overcome many of the problems associated with the 2D approaches, and would complement, if not better, some of the initial 3D efforts mentioned above. If validated, it will offer the advantages of the implanted markers for quantifying myocardial function, is non-invasive in nature, and can follow a much denser field
The organization of the rest of the paper is as follows: First, we will briefly describe the experimental setup for the in vivo acute infarct animal model and the 3D imaging techniques. Then, the shape-based image analysis methods, particularly our new developments in surface tessellation, curvature estimation, point-wise non-rigid motion tracking, initial quantitative measures of LV motion and thickening, and 3-D visualization techniques, will be discussed. And finally, initial experimental results from real image sequences will be presented, and future research directions will be stated.