Automatic Segmentation of Different Pathologies from Cardiac Cine MRI Using Registration and Multiple Component EM Estimation

In this paper, we develop a framework for the automatic detection and segmentation of the ventricle and myocardium from multislice, short-axis cine MR images. The segmentation framework has the ability to deal with large shape variability of the heart, poorly defined boundaries and abnormal intensity distribution of the myocardium (e.g. due to infarcts). We integrate a series of state-of-the-art techniques into a fully automatic workflow, including a detection algorithm for the LV, atlas-based segmentation, and intensity-based refinement using a Gaussian mixture model that is optimized using the Expectation Maximization (EM) algorithm and the graph cut algorithm. We evaluate this framework on three different patient groups, one with infarction, one with left ventricular hypertrophy (both are common result of cardiovascular diseases) and another group of subjects with normal heart anatomy. Results indicate that the proposed method is capable of producing segmentation results that show good robustness and high accuracy (Dice 0.908 +/- 0.025 for the endocardial and 0.946 +/- 0.016 for the epicardial segmentations) across all patient groups with and without pathology.