Computer-assisted determination of left ventricular endocardial borders reduces variability in the echocardiographic assessment of ejection fraction

Left ventricular size and function are important prognostic factors in heart disease. Their measurement is the most frequent reason for sending patients to the echo lab. These measurements have important implications for therapy but are sensitive to the skill of the operator. Earlier automated echo-based methods have not become widely used. The aim of our study was to evaluate an automatic echocardiographic method (with manual correction if needed) for determining left ventricular ejection fraction (LVEF) based on an active appearance model of the left ventricle (syngo AutoEF, Siemens Medical Solutions). Comparisons were made with manual planimetry (manual Simpson), visual assessment and automatically determined LVEF from quantitative myocardial gated single photon emission computed tomography (SPECT). 60 consecutive patients referred for myocardial perfusion imaging (MPI) were included in the study. Two-dimensional echocardiography was performed within one hour of MPI at rest. Image quality did not constitute an exclusion criterion. Analysis was performed by five experienced observers and by two novices. LVEF (%), end-diastolic and end-systolic volume/BSA (ml/m2) were for uncorrected AutoEF 54 +/- 10, 51 +/- 16, 24 +/- 13, for corrected AutoEF 53 +/- 10, 53 +/- 18, 26 +/- 14, for manual Simpson 51 +/- 11, 56 +/- 20, 28 +/- 15, and for MPI 52 +/- 12, 67 +/- 26, 35 +/- 23. The required time for analysis was significantly different for all four echocardiographic methods and was for uncorrected AutoEF 79 +/- 5 s, for corrected AutoEF 159 +/- 46 s, for manual Simpson 177 +/- 66 s, and for visual assessment 33 +/- 14 s. Compared with the expert manual Simpson, limits of agreement for novice corrected AutoEF was lower than for novice manual Simpson (0.8 +/- 10.5 vs. -3.2 +/- 11.4 LVEF percentage points). Calculated for experts and with LVEF (%) categorized into < 30, 30-44, 45-54 and > or = 55, kappa measure of agreement was moderate (0.44-0.53) for all method comparisons (uncorrected AutoEF not evaluated). Corrected AutoEF reduces the variation in measurements compared with manual planimetry, without increasing the time required. The method seems especially suited for unexperienced readers.