Automatic computation of left ventricular ejection fraction from spatiotemporal information in cine-SSFP cardiac MR...

To clinically validate an algorithm that automatically computes left ventricular (LV) ejection fraction (LVEF) using a priori geometric and intrinsic spatiotemporal information from cine steady-state free precession (SSFP) MR images.

The algorithm was evaluated in 64 subjects (21 healthy volunteers and 43 patients, LVEF 19-71%). Bland-Altman analyses were performed on short-axis slices subdivided into three sections (basal, midcavity, and apical) to assess the impact of morphologic variations on LVEF computation.

The automated algorithm delineated the clinically applicable endocardial boundary in 1011 of 1078 short-axis slices (94%). The bias (mean difference) values computed with clinically unusable contours replaced with hand-drawn equivalents were small for the LV end-diastolic volume (LVEDV, <11 mL/7%), end-systolic volume (LVESV, <7 mL/11%), and LVEF (<1.2%). Moreover, these values were within the limits of interobserver and intraobserver variability of experienced observers (LVEDV, <13 mL/8%; LVESV, <12 mL/17%; and LVEF, <5%). In the end-diastolic phase, the limits of agreement (bias +/- 1.96 SD of difference) were small (<5% LVEDV) in all sections. However, in the end-systolic phase, the limits of agreement were larger for the midcavity (<21% LVESV) and apical (<11% LVESV) slices.

This data-driven algorithm can estimate LVEDV, LVESV, and LVEF with a bias that is comparable to the interobserver and intraobserver variability of experienced observers.