Estimation of left ventricular cavity area with an on-line, semiautomated echocardiographic edge detection system.

Publisher Website
Google Scholar
Abstract
BACKGROUND Automated edge detection of endocardial borders in echocardiograms provides objective, reproducible estimation of cavity area; however, most methods have required off-line analysis. A recently developed prototype echocardiographic imaging system permits real-time automated edge detection during imaging and thus, the potential for measurement of cyclic changes in cavity area and the assessment of left ventricular function on-line. Our purpose was to compare measurements of endocardial area manually traced from conventional echocardiograms with those obtained with the real-time automated edge detection system in normal subjects. METHODS AND RESULTS Two training sets of images were used to establish optimal methods of gain setting; the settings were then evaluated in a test set of images. In the high-gain training group (n = 8 subjects, 119 images), gain settings were adjusted sufficiently high to display at least 90% of the endocardial border. Manually drawn and real-time area measurements correlated at r = 0.92, but manually drawn areas were underestimated by computer. In the low-gain training group (n = 7 subjects, 104 images), gain settings were adjusted sufficiently low to avoid cavity clutter despite the presence of dropout of endocardial edges. Manually drawn and real-time areas again correlated (r = 0.79), but manually drawn areas were overestimated by computer. In the intermediate-gain test group (n = 7 subjects, 105 images), gain settings were balanced between maximal endocardial definition (greater than or equal to 90%) and minimal cavity clutter (less than or equal to 1 cm2). Manually drawn and real-time areas correlated at r = 0.91 for the group, and r ranged from 0.94 to 0.99 in individual subjects. Interobserver variability was 9.5% for manually traced areas and 10.6% for real-time area measurements. CONCLUSIONS Real-time on-line automated edge detection provides accurate estimation of manually drawn cavity areas. Although the method is gain dependent, measurements are reproducible. The system should have clinical application in settings in which measurements of left ventricular function are important.