Efficient method for analyzing MR real-time cines: Toward accurate quantification of left ventricular function.

Abstract

To develop and assess an efficient method to identify end-expiratory end-diastolic (ED) and end-systolic (ES) images for accurate quantification of left ventricular (LV) function in real-time cine imaging. ECG-free free-breathing real-time cine imaging was performed on short-axis slices of thirteen healthy volunteers at 3 Tesla. K-means cluster segmentation was applied to delineate the endocardial contour, from which the LV centroid and cavity area were determined in each cine image. LV centroid displacement along the superior-inferior direction was filtered to extract respiratory motion in each slice. End-expiratory ED and ES images were then identified and used for LV function quantification. Accuracy was evaluated with that measured from the slice-matched standard ECG-gated breathhold segmented cines using two-tail paired Student's t-tests, linear regression analyses and Bland-Altman plots. Intra- and interobserver variability was calculated for each imaging technique. Qualitatively, end-expiratory ED and ES images identified with the proposed method agreed with those determined by frame-by-frame visual inspection in 97.5% of cases (P > 0.05). Quantitatively, good agreement of LV function indices between the real-time cine and the standard segmented cine was observed with averaged differences of 0.1 ± 0.9 g for myocardium mass, -0.3 ± 1.0 mL for ED volume, 0.2 ± 1.2 mL for ES volume, -0.2 ± 1.3 mL for stroke volume, and -0.3 ± 0.8% for ejection fraction. Paired LV function values exhibited strong correlation (r ≥ 0.96) and no significant difference (P > 0.05). The real-time cine and the standard segmented cine showed similar intra- (1.2-3.3% and 1.1-2.8%, respectively) and interobserver variability (2.6-6.9% and 1.8-4.8%, respectively) with all P-values > 0.05. All the variability was comparable with published results. Compared with the laborious frame-by-frame visual inspection, as conventionally adopted, the proposed method is efficient in analyzing real-time cines for the accurate quantification of LV function without excessively manual interactions.