MODEL-BASED QUANTIFICATION OF MYOCARDIAL PERFUSION IMAGES FROM SPECT

Abstract

A system for quantitative analysis of myocardial perfusion tomograms is proposed. The system starts with an automated delineation of the total left ventricle, including possible perfusion defects, to determine the mass and shape of the myocardium. Next, polar maps or bulls-eyes are computed from the delineation, which can then be compared to reference bulls-eyes to detect perfusion defects. The proposed system differs in three main aspects from currently available bulls-eye algorithms. First, radial slices are used rather than short-axis slices. In this way three-dimensional gradient information is retianed, in particular near the base and the apex of the left ventricule. Moreover, the reproducibility of this method is expected to be superior, since the interactive selection of short axis slices through the left ventricle is eliminated. Second, the left ventricle is automatically delineated using a flexible computer model in order to obtain higher reproducibility. The resulting delineation contains both mass and shape information. Third, in addition to the classic count rate bulls-eye, a mass bulls-eye is computed, which contains the myocardial mass corresponding to each bulls-eye pixel. Analysis of the count rate bulls-eye reveals perfusion defects, the quantification of the defects is carried out with the mass bulls-eye.