Automatic Extraction and Matching of Neonatal Cerebral Vasculature
Hui Xue, Christina Malamateniou, Joanna M. Allsop, Laxmikarthika V. Srinivasan, Joseph V. Hajnal, Daniel Rueckert · 2006
This paper presents a method for extracting and matching the neonatal cerebral vasculature from magnetic resonance angiography-time of flight (MRA-TOF) images. The vasculature is first extracted using a fully automatic version of the ridge traversal algorithm. Extracted vessel segments are then connected iteratively to compose a vessel tree that is automatically labeled. After this, an indirect vasculature registration method is used to recover global deformation between two vessel trees and vessel matching is performed by comparing a cost function measuring average spatial distance between two vessel branches. This process starts from roots of the trees and continues until the leaf branches are reached. The robustness and accuracy of vessel matching are improved by performing subtree registration using a robust bootstrap extension of the iterative closest point (ICP) algorithm. Experiments on data from neonatal brains show the effectiveness of the proposed vessel extraction and matching methods for analysis arteries subject to change resulting from growth and development