Multi frame guided local search for semiautomatic endocardial contour estimation in echocardiography sequences

We present a Guided Local Search (GLS) algorithm in a multi-frame environment for the detection of endocardial contours in echo cardiographic image sequences. The proposed method requires minimal user intervention at the first frame of the image sequence for specifying the candidate points of the contour. The initial contour is approximated by fitting an ellipse in the region defined by the points specified. First level of regularization on the contour is made by radial gradient search incorporating both magnitude and direction. The second level of regularization makes use of the GLS algorithm. The GLS with features and constraints for solution space and penalty terms overcome the optimization being trapped in local minima. The proposed method also incorporates the temporal information from the adjacent frames during the regularization process. The subsequent frames in sequence are initialized by the final contour of the previous frame. The algorithm developed was applied to 20 data sets over a cardiac cycle and the results were validated by comparing computer generated boundaries to those manually outlined by one expert. The maximum error in the contours detected was ±3.1mm. The application of guided local search algorithm proposed in this paper offers a fast and efficient method for semiautomatic segmentation of heart chambers in 2D echocardiography sequences.