Computational strategy for the segmentation of the aortic annulus in cardiac computed tomography images

Abstract

The purpose of this research is to segment the aortic annulus, present in cardiac computed tomography images, from a computational strategy generated using global similarity enhancement, vector least squares support machines and a segmentation technique named region growing. This enhancement is obtained by applying the following steps: a) Obtain an image of similarity by calculating the absolute value of the arithmetic subtraction that considers the original image and an image of contours. The image of contours is obtained by processing the original image with a filter based on the magnitude of the gradient. b) Is is processed with a Gaussian filter, generating a smoothed similarity image. On the other hand, considering the smoothed similarity image, the least squares support machines are used both to construct two cutting planes (that isolate the aortic artery) and to detect the coordinates of a seed voxel. In order to produce the morphology of the aortic valve annulus, the seed voxel initializes the technique of growth of regions during the segmentation process. From this morphology, some useful quantitative descriptors are calculated for the aortic ring characterization.