Automatic human brain vessel segmentation from 3D 7 Tesla MRA images using fast marching with anisotropic directional prior

Accurate 3D models of the human brain vessels can greatly help to diagnose serious diseases. Such models can be constructed by segmentation of 3D MRA images, especially the recently introduced high resolution 7T MRA. We propose a new two-step approach for fully automatic segmentation of 7T MRA images of the human cerebrovascular system. First, a 3D model-based approach is applied to segment thick vessels and most parts of thin vessels. Then, the missing vessel parts, which are caused by low contrast and high noise, are completed using a novel fast marching approach with anisotropic directional prior. An evaluation of our approach and a comparison with two previous approaches have been conducted using high resolution 3D 7T MRA images.