The quantification of left ventricular (LV) regional function from diagnostic images permits clinically important measurements to be made that are crucial for managing patients with ischemic heart disease. However, such measurements have been hampered by limitations in conventional imaging and image analysis methodology. These limitations have prohibited the accurate regional assessment of myocardial injury and/or the prediction of vessel patency, and include the facts that: 1.) most conventional imaging methods provide two-dimensional (2D) temporal slice or projection sequences, which cannot possibly permit viewing or analysis of the true motion of the heart, which rigidly moves and deforms in a three-dimensional (3D) space. 2.) Most approaches to regional image-based analysis of LV function: a.) make gross and restrictive assumptions about the general direction of LV motion or thickening (e.g. towards a center of mass) and b.) utilize only the end diastolic (ED) and end-systolic (ES) image frames, ignoring the fact that the LV actually goes through a temporal wave of contraction and the asynchrony of surface motion or LV thickening from region to region may be indicative of ischemia. The proposed research will use advanced techniques from computer vision to analyze 3D cardiac image sequences in order to more accurately estimate regional LV function. Our methodology makes use of mathematical optimization models related to the motion of 3D elastically deformable objects, an approach that is much more adaptable to the nonlinear, non-rigid regional motion of the LV than those based on the restrictive assumptions mentioned above. The approach will estimate and track the trajectories of motion of a dense field of points that sample the LV endocardial and epicardial surfaces at 1 point in time, over the entire cardiac cycle. In addition, these trajectories will be used to create quantitative measures of endocardial surface motion and LV thickening. the system will be validated using computer simulations of linearly deformable objects and image data acquired from acute dog studies. The acute dogs will have image-distinguishable markers sewn to the LV wall, and will be used not only for validation, but also to test the algorithm-derived measures' ability to predict the location and extent of myocardial injury. This testing will use data from two different imaging modalities, 4D gated/cine Magnetic Resonance Imaging and 4D real-time cine Computed Tomographic Imaging (using data from the Dynamic Spatial Reconstructor). Ultimately, our goal is to better distinguish and characterize the spatial/temporal extent and function of regions of ischemia and infarction in the LV wall, as well as gain insight into LV post-MI remodeling issues.
