Using morphological and clustering analysis for left ventricle detection in MSCT cardiac images

In this paper, an unsupervised approach based on non-linear filtering and region growing techniques to obtain the endocardial surface is proposed. The filtering stage is performed using mathematical morphology operators in order to improve the left ventricle cavity information in multi slice computerized tomography images. A seed point located inside the cardiac cavity is used as input for the region growing algorithm. This seed point is propagated along the image sequence to obtain the left ventricle surfaces for all instants of the cardiac cycle. The method is validated by comparing the estimated surface with respect to left ventricle shapes drawn by a cardiologist. The average error obtained was 1.38 mm.