Endocardium extraction using selective ensemble learning method in echocardiogram

A novel endocardium extraction method was proposed for echocardiogram based on the selective ensemble learning algorithm.Firstly,the region of interest(ROI) was initialized by incorporating the shape-based global prior. Secondly,various features were extracted for each candidate pixel in the ROI.The local features were learned and tested through the selective ensemble learning algorithm to obtain the fuzzy results.Lastly,these pixels' values were conveniently used in the snake model as the outer energy to eliminate the ambiguity and obtain final contours.This method was evaluated and compared with other contour extraction methods for real echocardiograms.The results demonstrated that the mean absolute difference and percentage of area overlap have decreased by 2.56 pixels and increased by 17.80% respectively.The proposed method was insensitive to speckle and initial contour.Our segmented boundaries were very close to the manually-segmented ones.It was indicated that our method had a better overall performance than existed ones.