Image processing:

Each image was pre-processed to reduce noise using anisotropic diffusion filtering [7]. A segmentation based on signal intensities and voxel connectivity [8,9] was then performed using one of the MR series (SPGR and MR angiogram). Imperfections in the 3D objects were corrected by manual editing. From these images, 3D models of the skin, brain, vessels, focal lesion and ventricles were reconstructed using the marching cubes algorithm and a surface rendering method [8,9,10]. These objects were then integrated and displayed on a computer workstation (Ultra-1; Sun Microsystems, Inc.) with a graphic accelerator (Creator 3D; Sun Microsystems Inc.) and 3D display software (LAVA; Sun Microsystems, Inc.). Each object thus may be rendered partially or completely transparent and oriented according to the viewer's choice. T2-weighted images and MR angiogram images can be registered to the base series (SPGR) using a multi-modal volume registration method by maximization of mutual information [36]. Each object may be individually colored to facilitate visualization.