Fusion of Resampled 3D MR Images for Geometric Modeling of Blood Vessels

Accurate quantitative information about blood vasculature depicted in 3D MR images has a great importance in diagnosis of many vascular diseases. In this work, the accuracy of vessel lumen walls reconstruction is studied. A 3D printed model of two MRI time-of-flight arterial branches is manufactured. Next, it is submerged in water and T2-weighted MR images are acquired in coronal, transversal and sagittal thick slice orientation. Then, MR anisotropic and fused (resampled and averaged) images are segmented with the use of Level-Set (LS) and centerline-radius (CR) algorithms. Surfaces of 3D branch models are constructed, and errors of their estimated radii are evaluated. We have demonstrated that LS approach may not be capable of segmenting thin vascular branches in anisotropic-voxel images. After image resampling, the LS segmentation restores the object surface, but with significant staircase distortion. Thanks to the knowledge about the assumed vessel shape, incorporated with the CR method, it produces accurate estimates of the branch radius and smooth model surface, even in the case of anisotropic voxel images.