Left ventricle segmentation with mixture of Gaussian active contours

Cine MRI is an imaging technique for visualizing the left ventricle wall motion in a heart beating cycle. The paper proposed a method to semi-automatically segment the left ventricle (LV) boundaries from a stack of cine MRI and build a LV mesh. The different intensities in the MRI are modeled with the mixture of Gaussian method. The thresholds between them are semi-automatically determined by K-means initialization and the expectation maximization (EM) method. The spatial constrain is added into the model with active contour models, which fit explicit contours on the boundary of the area of interest. We propose a marching cube alike initialization method of the active contour models. A 3D mesh is built from a stack of contours and smoothed afterwards. Stroke volume, myocardium mass and ejection function can be estimated from the LV boundary mesh and the endocardium mesh.