Motion Estimation from Tagged MR Images and Application to Cardiac Strain Evaluation and Functional Analysis of Ocular Tissues

Tagged MRI is a magnetic resonance imaging technique that has been widely applied to tissue functional evaluation, especially the myocardium. Unlike traditional anatomical imaging methods, tagged MRI spatially encodes the motion of underlying tissue by modulating the longitudinal magnetization periodically before deformation. The visual appearance is dark strips or grids that follow along with the tissue motion. Therefore, by investigating the apparent changing of the tag pattern, one can estimate the deformation pattern and mechanical properties of the target tissue. Previous motion estimation methods either manually or automatically reconstruct the deformation field from tagged MRI. However, most of them have limitations such as low computational efficiency or accuracy, or the inability to reconstruct a 3D deformation. In this dissertation, we develop three motion estimation methods that reduce manual intervention rate and computation time, while retaining a relatively high accuracy. The first introduced method improves a previous unwrapped phase-based method by automatically optimizing the branch cut configuration using simulated annealing. Then a quality-guided phase unwrapping follows after the application of branch cuts to unwrap the harmonic phase image. The unwrapped phase values provide displacement measurement and can be used to reconstruct a 3D deformation field. Although this method reduces manual intervention greatly compared to the previous manual method, it is time consuming. Then, we introduce a second method, which unwraps the harmonic phase image using integer optimization with graph cuts. The resulting unwrapped phase is both spatially and temporally smooth, as we assume the underlying deformation is continuous and smooth. We also use a dynamic model with a Kalman filter to improve the performance on biventricular studies. This method largely reduces the computation time and is highly automated. The third