Improved GVF Based Left Ventricle Segmentation from Cardiac MR Images Using Radial B-Snake Model

Segmentation of the left ventricle (LV) is a hot topic in cardiac magnetic resonance (MR) images analysis and still remains an open issue. In this paper, we propose a novel method, radial B-Snake model based on improved gradient vector flow (GVF), to segment LV automatically. Due to the left ventriclepsilas circle-like shape prior in short-axis view, the region of interest (ROI) of LV could be transformed into polar coordinates by taking its centroid as an origin. In the transform image, myocardium appears to be a horizontal band which enables snake to evolve towards 1D radial direction instead of 2D image plane simplifying the model and a line-shape constraint is incorporated to maintain the global features. For endocardium extraction, two-phase improved GVF strategy is developed to overcome image noise, artifacts and papillary muscles. For epicardium segmentation, GVF remodeled skillfully reactivates the resultant internal contour forward to epicardial boundary successively. Several experiments are presented to demonstrate the effectiveness and robustness of the proposed method.