Automatic Segmentation of Short-Axis Cardiac MRI using a Biventricular Deformable Model with an Explicit Thickness Prior

Paul A. Yushkevich¹, Hui Sun¹, Federico M. Sukno², Catalina Tobon-Gomez², Hongzhi Wang¹, Alejandro F. Frangi²

This new approach overcomes some of the challenges in automatic cardiac MRI segmentation of by modeling the myocardium using a deformable model that explicitly describes the skeleton of the myocardium and its thickness. This geometrical model is coupled with an appearance model and statistical shape prior learned from training data. The proposed approach achieves high accuracy compared to manual segmentation on end-diastole MR images of patients with common pathologies.