An algorithm for localizing branch points of pulmonary vessels for nonrigid registration of the lungs
Hidenori Shikata, Zhiming Yin, Yongsup Park, Hiroko Kitaoka, Yoshinobu Sato, Takeshi Johkou, Hironobu Nakamura, Shinichi Tamura · Systems and Computers in Japan · 2004
Abstract Analyzing three‐dimensional image data based on time series plays an important role in analyzing the functions of human organs that function with periodic changes in their shapes. Our final goal is to analyze the breathing movement of the lungs by nonrigid registration of the volume data between different time phases with vascular branch points spanning the lungs as characteristic points and obtaining positional vectors of each point. This paper presents a method for accurately detecting the locations of the branch points using graph representations of the blood vessels as an essential technique for such analyses. The method proposed obtains graph representations of an object using a method previously published, obtains approximately a product set region by taking each branch as a cylinder by local distance transforms in the neighborhood of a branching point obtained from the graph representations, and takes the centroid of this region as the branching point. The accuracy and reproducibility of determination of the position of a branch point are very important for nonrigid registration using the branch points. The authors have shown the efficacy of the proposed method by constructing a model simulating a blood vessel system by combining the cylinders, evaluating the accuracy using the differences from the axial intersection points of the cylinders constituting the branch as a measure. The authors have confirmed experimentally that the proposed method detects a branch point at a branch point position defined without being significantly influenced by conditions of rotations etc. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(3): 24–36, 2004; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/scj.10478