Robust Segmentation for Left Ventricle Based on Curve Evolution

This paper presents a novel multi-resolution framework for the segmentation of left ventricle in echocardiographic images. This framework is based on curve evolution and nonlinear diffusion pyramid. At the low resolution, a statistical region-based model is applied to analyze the echocardiographic images and it is combined with a boundary-based model for the pre-segmentation. The pre-segmentation result is used to initialize the front for the high resolution. Meanwhile, a fast mathematical morphology-based method is used to pass the solution from low to high resolution. This method is competent to fast narrowband re-initialization. Furthermore, a local Snake model is used as an external constraint to optimize segmentation at the high resolution. Segmentation results of left ventricle images show that the multi-resolution segmentation method is accurate and robust.