Carotid Artery: Model Construction

Realistic arterial models are reconstructed from anatomical images using tubular deformable models. First the vessel axis of each arterial branch is interactively selected on cross-sectional views (or obtained from skeletonization algorithms). Then a tubular deformable model along each vessel is constructed and allowed to deform due to forces from the image intensity gradient. The resulting triangulations are then merged into a water-tight surface using an adaptive voxelization technique. This triangulated surface is used as
a support surface or geometric definition of the computational domain for finite element grid generation. Unstructured grids composed of tetrahedral elements are generated using an advancing front method. Element sizes can be automatically specified from the surface curvature using adaptive background grids. A model constructed from magnetic resonance angiography (MRA) images is shown below.
 
 

Physiologic flow conditions are derived from phase-contrast magnetic resonance (PC-MR) measurements of blood flow velocity. Flow rate curves along each arterial branch are obtained by integration of the velocity profile over the cross section of the vessel. The vessel lumen is defined manually, or segmented from the magnitude images, or automatically segmented from the phase images using velocity cross-correllations. Pulsatile velocity boundary conditions are typically imposed using a supperposition of Womersley profiles for each Fourier mode.
 

The methodology described above was applied for the reconstruction of numerous normal and diseased carotid arteries from different imaging modalities including contrast-enhanced magnetic resonance angiography, CT angiography and 3D rotational angiography. The pictures below show some of the reconstructed models. These images illustrate the large anatomical variability among individuals. Since the flow dynamics strongly depends on the vessel shape, subject-specific models yield greater insight into arterial hemodynamics than computational models based on idealized geometries. Furthermore, patient-specific models are required for improved diagnosis and treatment planning.