The present invention relates to modeling of vascular structures, and in particular to extracting a vessel tree from medical-images of vascular anatomy. A respective method comprises among others the steps of providing an image of at least one vessel, obtaining multiple measurements from the image for a first point of the image, and fitting a four-dimensional tensor to the measurements. Based on said four-dimensional tensor fitted to the measurements, a vessel direction in a vessel tree is determined and a model of the vascular structures is generated based on at least the determined vessel direction.
