Adaptive enhancement of cardiac magnetic resonance (MR) images

This thesis presents a computational framework for the enhancement of magnitude magnetic resonance (MR) images of the heart. It addresses two sources of degradation of image quality from the perspective of an algorithmic procedure that seeks to recover the anatomical structures of the organ of interest: (1) the effect of image noise and other confounding features that obscure the relevant information of these structures and (2) inadequate image contrast. An analysis of the signal properties of these confounding features leads to the wavelet-based approach presented in this thesis. An enhancement framework, that is adaptive to both image structures and signal intensity, is developed to attenuate image noise, to remove the confounding features present in the low-intensity regions and to enhance image contrast. The framework is implemented and evaluated on both synthetic and medical test cases with strong supporting evidence for the hypothesis of this thesis.