A Fast Convex Optimization Approach to Segmenting 3D Scar Tissue from Delayed-Enhancement Cardiac MR Images

Abstract

Abstract. We propose a novel multi-region segmentation approach through a partially-ordered Potts (POP) model to segment myocardial scar tissue solely from 3D cardiac delayed-enhancement MR images (DE-MRI). The algorithm makes use of prior knowledge of anatomical spatial consistency and employs customized label ordering to constrain the segmentation without prior knowledge of geometric representation. The proposed method eliminates the need for regional constraint segmentations, thus reduces processing time and potential sources of error. We solve the proposed optimization problem by means of convex relaxation and introduce its duality: the hierarchical continuous max-flow (HMF) model, which amounts to an efficient numerical solver to the resulting convex optimization problem. Experiments are performed over ten DE-MRI data sets. The results are compared to a FWHM (full-width at half-maximum) method and the inter-and intra-operator variabilities assessed.

High-resolution 3-dimensional late gadolinium enhancement scar imaging in surgically corrected Tetralogy of Fallot: clinical feasibility of volumetric quantification and visualization

“…Importantly, phase reconstructed 2D LGE images were used for this analysis as (in our clinical experience) this provides improved visualization of scar within the RV wall. For 3D LGE scar segmentation, a novel, semi-automated software that employs a Hierarchical Max Flow algorithm was employed, as previously described [ 18 ]. This method uses a brush tool to obtain a single sampling of four distinct regions (Figure 2 ); normal myocardium, blood pool, scar and background.…”

Hierarchical max-flow segmentation framework for multi-atlas segmentation with Kohonen self-organizing map based Gaussian mixture modeling

“…However, these models have difficulty in managing multi-region segmentation problems in which several regions have individual regularization requirements not represented by a single smoothness term. This lack of topological knowledge has lead to the development of more nuanced max-flow segmentation models such as Ishikawa models (Ishikawa, 2003;Bae et al, 2011;Rajchl et al, 2012) although said models are constrained to segmentation problems in which the relationships between objects can be expressed using a full-ordering. This constraint poses difficulty for the segmentation of anatomy in which the part/whole relationships cannot be defined as such.…”

A variational method for scar segmentation with myocardial contour correction in DE-CMR images

“…probabilistic approach [3]. More elaborate proposals include the use of advanced segmentation methods [4] or postprocessing the outcome to remove false scars arising from errors in the location of the myocardial contours [5]. The latter, however, does not modify the location of the myocardial borders.…”