Contribución a técnicas de procesado de imagen relacionadas con ecuaciones en derivadas parciales para la ayuda al diagnóstico de cardiopatías mediante resonancia magnética de realce tardío

Abstract

Late Gadolinium Enhancement Cardiac Magnetic Resonance (LGE-CMR) is an imaging modality that allows for the identification of damaged tissue (scar or fibrosis) within the myocardium, since more gadolinium contrast accumulates on it due to perfusion defects. Thus, it is an important diagnostic tool in the context of pathologies such as ischemic or hypertrophic cardiomyopathy.Damaged tissue segmentation methods often start from a segmentation of the myocardium, given that the damaged tissue and the blood intensity values in LGE-CMR images are very similar and the boundaries between these regions are often difficult to detect. This myocardial segmentation may be manually delineated or computed from other imaging modality, such as CINE-CMR, and then registered to the LGE-CMR image. However, misalignment of the myocardial contours can lead to some of the adjacent tissue to be misclassified. The ratio of the myocardial thickness covered by scar (scar transmurality) has prognostic value for the outcome of reperfusion therapy in patients of ischemic cardiomyopathy. The scar transmurality is computed on short axis slices using 2D methods. However, the true point-to-point correspondences between the endocardium and the epicardium are not, in general, restricted to a short-axis plane.The main contribution of the present work is a method to compute myocardial thickness and dense scar transmurality maps based on a partial differential equation and implemented numerically using a multi-stencil scheme. The method has been tested using analytical functions and three different image banks of real LGE-CMR images. We also propose a multimodal segmentation method when both LGE-CMR and CINE-CMR images are available along with a CINE-CMR myocardial segmentation, which allows for small displacements of the endocardial and epicardial contours. It is a modification of an existing variational method that employs a novel expectation-maximization algorithm to estimate the probability distribution parameters using a log-likelihood function that incorporates both LGE-CMR and CINE-CMR images and the CINE-CMR myocardial segmentation. We also studied the influence of the chosen stencil set on the results of the Multi-Stencil Fast Marching method. A new version using a centered second-order finite difference scheme for the partial derivatives was also proposed.