Left Ventricle mass extraction utilizing a multi-step probabilistic approach

In this paper, a fully automated multi-step approach for segmenting the Left Ventricle (LV) chamber in echocardiography images is proposed. A preprocessing step is applied to remove the dark background and find the seed point inside the LV chamber, eliminating the specialist intervention for identifying the seed point to initialize the segmentation process. The Rayleigh-Gaussian Mixture model is used to statistically differentiate the blood from the myocardial regions in Echocardiography images. An ellipse inside the LV chamber is then automatically fitted by utilizing the active ellipse model. The resultant ellipse forms the initial contour of the modified B-Spline Snake algorithm that evolves iteratively to outline the inner boundary of the LV chamber. Finally, the Markov Random Field (MRF) algorithm is utilized to extract the LV myocardium. The boundary specified by the B-spline snake is involved in reducing the number of MRF sites. The detected myocardium accuracy for all images is calculated using the dice´s coefficient. Our results demonstrate that the proposed method is more reliable and overcomes the common problems in segmentation of Echocardiography images such as speckle noise and large gaps in contrast with previous works.