929-64 Global Cardiac Function Using Active Contour Models with Fast Breath-hold MRI

Fast MR imaging techniques now permit rapid acquisition of gated cardiac images spanning the entire ventricle through most of the cardiac cycle. Quantitative global function analysis requires extraction of the myocardial borders in the images. However, manual analysis is tedious and time consuming while fully automated techniques are limited because of image artifacts and poor blood/myocardial contrast in certain regions, especially with fast-imaging techniques. We have developed a fast, interactive, semi-automated method for defining the myocardial borders. We present the initial results of this method using MR images of static phantoms, dog hearts, and normal volunteers. Ten healthy adult volunteers (ages 18–34) were imaged with a clinical 1.5 Tesla MRI system (GE) using a fast gradient-echo ECG-gated pulse sequence which acquired 8–12 cardiac phases from end-diastole (ED) past end-systole during a breath-hold. 12–15 short-axis (SA) images were acquired from the pulmonary valve to the LV apex. The imaging parameters were: FOV = 24–;26 cm, flip angle = 33°, slice thickness = 6 mm, slice spacing = 2 mm, TR/TE = 7.2/2.6 msec. Analysis was done with our customized software (SPAMMVU) using our contour tracking routine. This method models each contour as an analytic function (with an associated effective strain “energy”) that is attracted to edges in the image, but can be manipulated by the user to outline regions with poor contrast or to avoid artifacts. Internal stiffness of the model results in smoothing and fast outlining. Typical analysis takes less than 20 minutes vs. 1–2 hours for a manual method. Volumes were computed using Simpson's rule. Measured volumes of static phantoms had less than 5% error compared to truth. Errors in calculated LV mass of dog hearts by MRI were less than 10% compared with actual ex vivo weight. LV mass of normal volunteers was determined at ED and ES, Linear regression analysis of ED LV mass vs. ES LV mass yields an excellent correlation (R 2 of 0.94). A similar analysis of LV stroke volume (SV) vs RV SV yields an R 2 value of 0,85. Our semi-automated contour extraction program provides a fast, reproducible, and accurate method for quantifying cardiac global function from MR images.