Fully-Automatic Segmentation of Cardiac Images Using 3-D MRF Model Optimization and Substructures Tracking

Abstract

We present a fully-automatic fast method for heart segmentation in pediatric cardiac MRI. The segmentation algorithm is a two step process. In the first step a 3-D Markov random field (MRF) model is assumed for labeling the MR images into four intensity classes, the two of them corresponding to the blood pool areas. The intensity distribution of the four classes is estimated by an unsupervised method. In the second step the resulting regions are, maybe further segmented and, classified in the three main categories: blood pool, myocardium and background. The classification is obtained by tracking the cardiac substructures that can be clearly distinguished in detected specific slices. The whole process is driven by the data analysis and by generic models on 2-D regions and 3-D volumes, without a deformation model, which eventually might be fitted. The algorithm is evaluated on the HSVMR 2016 data set in Congenital Heart Disease.