Comparison of semi-automated scar quantification techniques using high-resolution, 3-dimensional late-gadolinium-enhancement magnetic resonance imaging

Document Type

Article

Publication Date

2-27-2015

Journal

International Journal of Cardiovascular Imaging

Volume

31

Issue

2

First Page

349

Last Page

357

URL with Digital Object Identifier

10.1007/s10554-014-0553-2

Abstract

© 2014, Springer Science+Business Media Dordrecht. The quantification and modeling of myocardial scar is of expanding interest for image-guided therapy, particularly in the field of arrhythmia management. Migration towards high-resolution, three-dimensional (3D) MRI techniques for spatial mapping of myocardial scar provides superior spatial registration. However, to date no systematic comparison of available approaches to 3D scar quantification have been performed. In this study we compare the reproducibility of six 3D scar segmentation algorithms for determination of left ventricular scar volume. Additionally, comparison to two-dimensional (2D) scar quantification and 3D manual segmentation is performed. Thirty-five consecutive patients with ischemic cardiomyopathy were recruited and underwent conventional 2D late gadolinium enhancement (LGE) and 3D isotropic LGE imaging (voxel size 1.3 mm3) using a 3 T scanner. 3D LGE datasets were analyzed using six semi-automated segmentation techniques, including the signal threshold versus reference mean (STRM) technique at >2, >3, >5 and >6 standard deviations (SD) above reference myocardium, the full width at half maximum (FWHM) technique, and an optimization-based technique called hierarchical max flow (HMF). The mean ejection fraction was 32.1 ± 12.7 %. Reproducibility was greatest for HMF and FWHM techniques with intra-class correlation coefficient values ≥0.95. 3D scar quantification and modeling is clinically feasible in patients with ischemic cardiomyopathy. While several approaches show acceptable reproducibility, HMF appears superior due to maintenance of accuracy towards manual segmentations.

This document is currently not available here.

Share

COinS