Background: We developed a new segmentation algorithm based on the invariance of the Laplacian (IL) to compute volumes and ejection fractions and compared these results with planar analysis and gradients by use of a standard algorithm (QBS).
Methods and results: Planar and single photon emission computed tomography blood pool acquisition was performed in 202 patients. Planar left ventricular ejection fraction (LVEF) was used as the gold standard, and single photon emission computed tomography images were processed by both 3-dimensional (3D) methods. Correlations between each 3D algorithm and planar methodology were as follows: r = 0.77 for QBS and r = 0.84 for IL. Mean LVEFs were 32.72% +/- 13.05% for the planar method, 32.32% +/- 15.98% for QBS, and 31.93% +/- 13.44% for IL (P = .16). Bland-Altman analysis closely demonstrated negligible systematic bias for both 3D methods. Standard errors of bias were comparable between methods (9.36% for QBS and 7.44% for IL, P = .48). Linear regression of the Bland-Altman bias revealed a slope significantly different from 0 for the QBS method (0.22 +/- 0.048, P < .0001) but not for IL (-0.032 +/- 0.0044, P = .47).
Conclusion: The new segmentation algorithm provides comparable results to QBS and planar analysis. However, with QBS, the difference in LVEF was correlated with the magnitude of LVEF, which was not found with the new algorithm.