Evaluation of the inter-observer cardiac chamber contour extraction versus a level set algorithm

Abstract

Segmentation of echocardiography images presents a great challenge because these images contain strong speckle noise and artifacts. Besides, most ultrasound segmentation methods are semi-automatic, requiring initial contour to be manually identified in the images. In this work, a level set algorithm based on the phase symmetry approach and on a new logarithmic based stopping function is used to extract simultaneously the four heart cavities in a fully automatic way. Then, those contours are compared with the ones obtained by four physicians to evaluate the performance, reliability and confidence for eventual clinical practice. That algorithm evaluation versus clinicians' performance is made using several metrics, namely Similarity Region, Hausdorff distance, Accuracy, Overlap, Sensitivity, and Specificity. We show that the proposed algorithm performs well, producing contours very similar to the physicians' ones with the advantage of being an automatic segmentation technique. The experimental work was based on echocardiography images of children. © 2011 Springer-Verlag.