In this paper, a geometric active contour for segmentation of the myocardium of the left ventricle in cardiac magnetic resonance (CMR) images is presented. The stochastic active contour scheme (STACS) and its variants, such as Li\\\'s method [12], are well-known frequently-used approaches for segmentation of the myocardium boundaries. However, they have significant difficulties with the image inhomogeneity due to using a region-based energy term established on the global Gaussian probability density function. It seems that localizing the region-based term of the energy functional is an effective solution to tackle the above problem. We enhance Li\\\'s method by substituting the region-based terms of endocard and epicard active contours with those of the local binary fitting (LBF) and local Gaussian distribution fitting (LGDF), respectively. Both LBF and LGDF belong to the category of patch-based active contours which can essentially handle the image inhomogeneity. Experimental results demonstrated that the proposed method provides significantly superior performance compared to Li\\\'s method in segmentation of the myocardium of the left ventricle.

Keywords