Fully automatic segmentation of left ventricular anatomy in 3-D LGE-MRI. (July 2017)

• Record Type: Journal Article
• Authors: Kurzendorfer, Tanja; Forman, Christoph; Schmidt, Michaela; Tillmanns, Christoph; Maier, Andreas; Brost, Alexander
• Is Part Of: Computerized medical imaging and graphics. Volume 59(2017)
• Page Start: 13
• Page End: 27
• Publication Date: 2017-07
• Subjects: Segmentation -- Left ventricle -- Late-gadolinium-enhanced magnetic resonance imaging; Diagnostic imaging -- Periodicals; Imaging systems in medicine -- Periodicals; Diagnosis, Radioscopic -- Data processing -- Periodicals; Diagnostic Imaging -- Periodicals; Imagerie pour le diagnostic -- Périodiques; Diagnostic imaging; Periodicals; Electronic journals; Electronic journals; 616.0754
• DOI: 10.1016/j.compmedimag.2017.05.001
• Languages: English
• ISSNs: 0895-6111

Abstract

Graphical abstract: Highlights: Fully automatic segmentation of the LV using 3-D LGE-MRI. Review of state-of-the-art methods. Comparison to gold-standard annotation and evaluation of inter-observer variability. Evaluation on patient data from two different LGE-MRI sequences at two clinical sites. Abstract: The current challenge for electrophysiology procedures, targeting the left ventricle, is the localization and qualification of myocardial scar. Late gadolinium enhanced magnetic resonance imaging (LGE-MRI) is the current gold standard to visualize regions of myocardial infarction. Commonly, a stack of 2-D images is acquired of the left ventricle in short-axis orientation. Recently, 3-D LGE-MRI methods were proposed that continuously cover the whole heart with a high resolution within a single acquisition. The acquisition promises an accurate quantification of the myocardium to the extent of myocardial scarring. The major challenge arises in the analysis of the resulting images, as the accurate segmentation of the myocardium is a requirement for a precise scar tissue quantification. In this work, we propose a novel approach for fully automatic left ventricle segmentation in 3-D whole-heart LGE-MRI, to address this limitation. First, a two-step registration is performed to initialize the left ventricle. In the next step, the principal components are computed and a pseudo short axis view of the left ventricle is estimated. The refinement of the endocardium and epicardium is performed in polar space. Prior knowledge for shape and inter-slice smoothness is used during segmentation. The proposed method was evaluated on 30 clinical 3-D LGE-MRI datasets from individual patients obtained at two different clinical sites and were compared to gold standard segmentations of two clinical experts. This comparison resulted in a Dice coefficient of 0.83 for the endocardium and 0.80 for the epicardium.