Contornos ativos geométricos e paramétricos aplicados à segmentação do ventrículo esquerdo em imagens de ecocardiograma

Abstract

The correct left ventricle identi cation in echocardiography images constitutes an im- portant tool to early diagnosis of heart diseases. However, limitations inherent to the im- age acquisition process, as low signal-to-noise ratio and presence acoustic shadows, make this identi cation a complex task. In this work, purposing to segment the left ventricle, geometric and parametric active contours are applied over echocardiography images. A new parametric active contours variation, the algebraic active contours method, and two kinds of level-set methods, one based on gradient and another based on regions, are tested and compared. In this work, four contributions to literature are proposed. One is the Hamiltonian ltering by a gaussian window in order to accelerate the contour movement. Another one is a signal function, that allows the contour to change automatically its ra- dial direction. Moreover, the algebraic active contours method conception itself is one of the contributions. Finally, the last one is the contours map superposition, that increases active contour's in uence area without accuracy loss. It is concluded that, although the geometric active contours capacity of splitting and merging itself independently of new parametrizations, the algebraic active contours method is, among the evaluated methods, the only one able to segment the left ventricle in echocardiography images conform to the gold-stardard provided by the specialists.