The clinical utility of automatic boundary detection for the determination of left ventricular volume: a comparison with conventional off-line echocardiographic quantification.

The aim of this study was to compare measurements of echocardiographic volume with an on-line automatic boundary detection imaging system with those of a conventional off-line method for routine clinical studies. Automatic boundary detection imaging shows promise as a rapid, on-line method for quantitating left ventricular volumes by echocardiography. However, there is little information about the role of automatic boundary detection for routine clinical studies. Ninety-seven patients with a variety of clinical diseases who were referred for clinical transthoracic echocardiographic evaluation were studied in apical four-chamber and two-chamber imaging planes. End-diastolic volume, end-systolic volume, and ejection fraction obtained with automatic boundary detection images were compared with those of conventional off-line analysis. Segmental endocardial definition and border tracking were evaluated on all automatic boundary detection images. Left ventricular end-diastolic volumes obtained by automatic boundary detection correlated well but were systematically under-estimated compared with off-line analysis for the apical two-chamber (r = 0.83; underestimation = 42 +/- 33 ml; p < 0.05) and four-chamber views (r = 0.83; underestimation = 43 +/- 31 ml; p < 0.05). Left ventricular end-systolic volumes also correlated well but were underestimated by automatic boundary detection for the apical two-chamber (r = 0.83; underestimation = 14 +/- 26 ml; p < 0.05) and four-chamber views (r = 0.83; underestimation = 18 +/- 24 ml; p < 0.05). Ejection fraction was not predicted accurately for the entire study population (n = 97). However, for patients with complete endocardial definition (n = 32), automatic boundary detection accurately predicted ejection fraction with no systematic error compared with manually traced images for both the apical two-chamber (r = 0.86; p < 0.05) and four-chamber (r = 0.82; p < 0.05) views. Segmental analysis of endocardial tracking revealed significantly better tracking of the septal and lateral walls compared with other regions (p < 0.05). End-diastolic and end-systolic volumes determined by automatic boundary detection correlate well but underestimate volume compared with conventional off-line analysis. However, ejection fraction compares favorably for the two methods when there is complete endocardial definition.