CARDIAC MR IMAGE SEGMENTATION WITH INCOMPRESSIBILITY CONSTRAINT
Yun Hang Zhu, Xenios Papademetris, James S. Duncan, Albert J. Sinusas · 2007
Automatic segmentation of the left ventricle (LV) from cardiac images remains an open problem. While current methods are already sufficient to outline endocardial (ENDO) surface automatically, these methods are problematic for finding reliable epicardial (EPI) surfaces. It is mainly due to the low myocardium/background contrast. In this paper, we propose a new algorithm that is motivated by the approximate incompressibility of myocardium during a cardiac cycle and takes it as an important constraint. We design in a probabilistic framework a deformable model that evolves according to the regional intensity distribution while maintaining the volume of myocardium. Experiments on 225 sets of volumetric cardiac MR images validate the accuracy and robustness of this method