Fully automated data acquisition, processing, and display in equilibrium radioventriculography

A fully automated data acquisition, processing, and display procedure was developed for equilibrium radioventriculography. After a standardized acquisition, the study is automatically analzyed to yield both right and left ventricular time-activity curves. The program first creates a series of edge-enhanced images (difference between squared images and scaled original images). A marker point within each ventricle is then identified as that pixel with maximum counts to the patient's right and left of the count center of gravity of a stroke volume image. Regions of interest are selected on each frame as the first contour of local maxima of the two-dimensional second derivative (pseudo-Laplacian) which encloses the appropriate marker point, using a method developed by Goris. After shifting the left ventricular end-systolic region of interest four pixels to the patient's left, a background region of interest is generated as the crescent-shaped area of the shifted region of interest not intersected by the end systolic region. The average counts/pixel in this background region in the end systolic frame of the origina study are subtracted from each pixel in all frames of the gated study. Right and left ventricular time-activity curves are then obtained by applying each region of interest to its corresponding background-subtracted frame, and the ejection fraction, end diastolic, end systolic, and stroke counts determined for both ventricles. In fourteen consecutive patients, in addition to the automatic ejection fractions, manually drawn regions of interest were used to obtain ejection fractions for both ventricles. The manual regions of interest were drawn twice, and the average obtained. For the right ventricle, the correlation between auto and average manual ejection fraction was 0.52; the correlation between the two manual ejection fractions was 0.88. For the left ventricle, the correlation between auto and average manual ejection fraction was 0.96; the correlation between the two manual ejection fractions was 0.91. Automated processing is essential for the accurate and reproducible assessment of left ventricular ejection fraction.