[Automatic contour detection in the 2-dimensional echocardiogram--studies in a general patient population].

In order to test the application of an endocardial contour finding algorithm in apical projections we examined 56 consecutive patients by 2-D echocardiography. According to the quality of the endocardial definition we graded the total patient population in 4 qualities, grade I being the best and grade IV being the worst echocardiographic image with multiple extracavitary and intracavitary artefacts. Following manual and automatic detection of LV endocardium in end-systole and end-diastole by 2 observers we calculated LV areas (cm2), volumes (ml), ejection fraction (%) and regional wall motion (% radial shortening and % area shrinkage) and determined the reproducibility of the two different endocardial definitions. In all echocardiograms with good endocardial visualisation (grade I and II; n = 31, 55%), in 9 out of 15 patients with grade III and 4 out of 10 patients with grade IV echocardiograms we could successfully apply the contour finding algorithm. The comparison between automatically and manually detected contours showed good correlations with small standard errors; however the reproducibility of data expressed by the interobserver variability (V) calculated on the basis of automatically found contours was significantly better as compared to the manually derived contours. (V manual contour = 4.8-8.5 versus automatic contours 0.63-0.68). The systolic volumes (manual versus auto) correlated with r = 0.93, SEE 8.3 ml; the end-diastolic volumes (manual versus auto) with r = 0.90, SEE 11.3 ml, the ejection fraction (manual versus auto) with r = 0.93, SEE 5.1%. The determination of regional LV wall motion demonstrated an agreement in 91% and a discrepancy in 9% of regions.(ABSTRACT TRUNCATED AT 250 WORDS)