On motion in dynamic Magnetic Resonance Imaging: Applications in cardiac function and abdominal diffusion
MRI is a well-established medical imaging technique with excellent tissue contrast and high spatial resolution, without the need for ionizing radiation. However, quantification and compensation of physiological motion during acquisition represent a major issue that must be considered for the development of robust biomarkers. This thesis focuses on the estimation and correction of motion in different MRI modalities, to provide robustness in the assessment of functionality and tissue composition of abdominal organs. The para- meters we are seeking to provide should be robust, reproducible, with reduced intra- and inter-observer variability and easy to visualize for a better radiological interpretation. Specifically, this Thesis focuses on two main tasks: (1) the characterization of the me- chanical properties and possible misfunctionalities of the myocardium and (2) the robust estimation of the apparent diffusion coefficient in the liver. For the former, we propose a methodology for the robust estimation of motion and strain, as well as a procedure for identifying the presence of fibrotic tissue and classifying the different aetiologies behind hypertrophic cardiomyopathy. For the sake of comprehensiveness, we have also introduced a thorough description of the harmonic phase techniques and an extensive analysis of the different strategies for robust motion and strain estimation in cardiac tagged magnetic resonance. In addition, a review of the most relevant features in cardiomyopathy screening and classification is carried out. About the latter, we propose a joint registration and estimation procedure for abdomi- nal diffusion weighted imaging. This approach provides a reproducible apparent diffusion coefficient estimation, which is less sensitive to noise and physiological motion during acquisition, two of the main issues in clinical imaging. The main contribution in this se- cond application is twofold: first, the inclusion of a groupwise registration methodology aimed at minimizing the residuals in the estimation; second, the proposal of filtering sta- ges to alleviate the influence of noise in diffusion parameter estimation, which may lead to spuriously biased estimates, especially in low signal-to-noise-ratio scenarios. The pro-posed study also evaluates the decrease in accuracy when the noise model is not properly accounted for.