Model driven interpretation of velocity encoded aortic flow images by means of Voronoi arrangement matrices

In this paper, we investigate the application of a representation for tomographic similarity to recognize vessels in aortic flow images. Due to the strict acquisition protocol, the spatial embedding of different vessels in aortic flow images is largely similar for different subjects. By modeling the vessel embedding of these images by means of Voronoi arrangements, the spatial arrangement of a 'training' image can be utilized to label unknown objects in target images by matching their arrangements. The arrangement matching method was tested on routinely acquired flow images of 12 subjects, in which four major vessels were drawn manually. In all cases, the arrangement matching labeled the vessels in the images correctly, though clear anatomic differences in size and shape were present. This shows that the arrangement metric is a powerful representation of the content of images, in which objects differ in size, shape and location, but of which the spatial embedding is similar.