Quantitative study of cardiac motion estimation and abnormality classification in emission computed tomography.
Quantitative description of cardiac motion is desirable to assist in detecting myocardial abnormalities from gated myocardial perfusion (GMP) emission computed tomography (ECT) images. While "optical flow" type of cardiac motion estimation (ME) techniques have been developed in the past, there has been no quantitative evaluation of their performance. Moreover, no investigation has been performed in terms of applying an ME technique to quantify cardiac motion abnormalities. Using the four-dimensional NCAT beating heart phantom with known built-in motion, the current work aimed at addressing the aforementioned two issues. A three-dimensional cardiac ME technique was developed to search for a motion vector field (MVF) that establishes voxel-by-voxel correspondence between two GMP ECT images. The weighted myocardial strain energy served as the constraint in the process to minimize the difference between one intensity image and the MVF warped other. We studied the convergence of the ME technique using different initial estimates and cost functions. The dependence of estimated MVF on the initialization was attributed to the tangential motion that is undetectable while not suppressed by the strain energy constraint. We optimized the strain energy constraint weighting using noise-free phantom images and noisy reconstructed images, the former against the known MVF and the later in the task of regional motion classification. While the results from the above two studies well coincide with each other, we also demonstrated that upon appropriate optimization the ME method has the capability of serving as a computer motion observer in separating simulated noisy reconstructed GMP SPECT images corresponding to hearts with and without regional motion abnormalities.
Tang, J; Segars, WP; Lee, T-S; He, X; Rahmim, A; Tsui, BMW
Volume / Issue
Start / End Page
Pubmed Central ID
Electronic International Standard Serial Number (EISSN)
Digital Object Identifier (DOI)