Improved modelling of the human cerebral vasculature

Abstract

This thesis presents a hybrid strategy to build an elliptical model of the human cerebral vasculature from the three dimensional (3D) Time of flight (TOF) magnetic resonance angiography (MRA) data. The hybrid strategy includes 3 steps. In the first step, a novel segmentation method based on locally adaptive thresholding is proposed to generate 3D binary data of cerebral vasculature, which offers the superiority in accuracy and extraction of finer distal vessels. In the second step, a centerline extraction method is developed based on the wave propagation of the marching spheres to automatically generate a circular centerline model of the vasculature from binary data and build a vascular tree structure by providing robust bifurcation detection. In the third step, elliptical measurement is performed by extending the affine medialness from the linear medialness to approximate the vessel cross sections with ellipses. The hybrid strategy has been applied on TOF MRA data sets of 1 volunteer, 21 patients and 1 physical cerebral vascular phantom. The results show that our hybrid strategy is a better choice to facilitate automation, provide robust bifurcation detection and generate an elliptical centerline model as a more general and accurate description of vessels.