Interpreting edge information for improved endocardium delineation in echocardiograms
Successful segmentation of the left ventricle in echocardiograms relies on strong edge responses to ensure that segmentation methods converge to the endocardial boundary. However, segmentation methods that do not interpret edge responses using local shape information or global context are often led astray by imaging artifacts. An extension to a boundary fragment model borrowed from computer vision literature is presented as a method for determining which edge responses contribute to the endocardial boundary. To demonstrate its applicability, the proposed method is applied to a data set composed of long-axis echocardiogram slices from five subjects. Results show that the process is effective at locating the endocardium and identifying the edge responses which correspond to the endocardial boundary.
