Automatic segmentation of the left ventricle from dual contrast cardiac MR data

Manual tracing of the blood pool from short axis cine MR images is routinely used to compute ejection fraction (EF) in clinical practice. The manual segmentation process is cumbersome, time consuming, and operator dependent. In this paper, we present an algorithm for the automatic computation of the EF that is based on segmenting the left ventricle by combining the fuzzy connectedness and deformable model frameworks. Our contributions are the following: 1) we automatically estimate a seed point and sample region for the fuzzy connectedness estimates, 2) we extend the fuzzy connectedness method to use adaptive weights for the homogeneity and the gradient energy functions that are computed dynamically, and 3) we extend the hybrid segmentation framework to allow forces from dual contrast and fuzzy connectedness data integrated with shape constraints. We compared our method against manual delineation performance by experienced radiologists on the data from nine asymptomatic volunteers with very encouraging results.