| MRI provides the opportunity for noninvasive monitoring of organ function, in particular, of the heart function. To quantitatively analyze the dynamic function of a heart from magnetic resonance images, various parts of the heart chambers must be segmented. In clinical studies, the segmentation task is often performed manually, which is tedious and time consuming. To expedite the process, an automatic image segmentation approach is desired. We develop in this thesis a novel stochastic active contour scheme (STACS) for automatic image segmentation, designed to overcome some of the unique problems in cardiac MR images such as problems with low contrast or turbulent blood flow. STACS incorporates stochastic region-based and edge-based information derived from the image, as well as shape priors of the heart and local properties of the contour, to form an energy functional to be minimized. The minimization algorithm solves a partial differential equation (PDE) that describes the contour evolution using the level set method. STACS includes an annealing schedule that balances dynamically the weight of the different terms in the energy functional.; In this thesis, STACS was applied to both synthetic and real cardiac MR images. It segments the heart from its surroundings such as the chest wall, and the heart structures (left and right ventricles and epicardium.) Three particularly attractive features of STACS are: (1) stochastic models for image textures, resulting in an ability to segment images with low texture contrast; (2) robustness to initial contour and noise because of edge and region-based information; and (3) incorporating shape priors of the heart, resulting in the ability to segment the undesired papillary muscles out. Finally, we assess quantitatively the performance of STACS by comparing, for a set 48 real cardiac MR images, STACS' automatically generated contours against manually-traced contours, or the "gold standard," using both area and edge similarity measures. This assessment validates STACS' results, demonstrating its very good and consistent segmentation performance. |
