[Computer-assisted contour detection in a complete heart cycle by levocardiography with adaptive raster tracking--validation of the method].
A method for computer-assisted edge detection in a full heart cycle of the left ventricular cineangiogram is presented. Left ventricular cineangiograms were acquired in the 30 degrees RAO position on cinefilms with 50 frames/s. A selected heart cycle was digitized with a spatial resolution of 512 x 512 pixels and a grey-value range from 0 to 255. In the first frame of the heart cycle a raster was calculated from contour points marked by an operator. The raster was automatically adapted to the left ventricular contour in the following frames. The contour points were determined on the adaptive raster by a gradient method. The automatic calculation of all contours of the systolic contraction was possible in 28 of 30 cineangiograms, operator interaction was necessary in two cineangiograms. The calculation of the diastolic contours requested operator interaction in 4 of 30 cineangiograms. The mean deviation of the calculated from the true contours ranged from -1.2 +/- 1.9 mm to 0.9 +/- 2.9 mm. The correlation coefficients between volumes calculated from the computer-determined and manually drawn contours ranged from r = 0.98 to r = 1.0 at beginning of systole, midsystole, endsystole, middiastole, and enddiastole with a standard error of the estimate between 2.8 ml and 5.2 ml. A repeated automatic determination of the contours after an independent remarking of the starting points resulted in a linear correlation of the calculated volumes with a correlation coefficient of r = 1.0 and a standard error of the estimate between 1.3 ml and 4.7 ml. The correlation coefficient for the ejection fraction was r = 0.99 with a standard error of the estimate of 1.9%.(ABSTRACT TRUNCATED AT 250 WORDS)