| A three-dimensional (3D) digital image can be formed by stacking a contiguous sequence of two-dimensional (2D) cross-sectional images. It can be used to analyze complex 3D objects. Many medical-imaging scanners have been developed to generate 3D images of the anatomy. Recently, high-resolution x-ray microcomputed tomography (micro-CT) scanners have been devised to image microvasculature and other structures significantly under 1mm in size. Five characteristics are common to the generated images. First, they are very large, ranging in size from tens to hundreds, even thousands of megabytes (MB). Second, voxels further down the tree or interior to the tree may appear dimmer than other tree voxels. This occurs because of inconsistent spreading of the contrast medium. Third, intensity values along the network tend to drop as one moves from the tree's root. This again arises from the contrast medium's inconsistent spreading. Fourth, the tree root may occur at any location within the image. Finally, degradations of tomographic images arise from reconstruction (streak) artifacts, noise, and partial-volume averaging. An important problem that arises in many 3D imaging scenarios is to analyze the characteristics of large, complex branching networks in a huge 3D digital image. This analysis can reveal, for example, the relationships between the cross-section area (CSA) of a branch and its distance to the root. Also, for medical applications, the physician can exploit the CSA calculation along branches to identify blocked or stenosed branches or to understand the distribution of blood in an organ. To accomplish such analysis, suitable 3D image-processing and visualization tools are needed. This thesis seeks to devise processing and visualization methods suitable for analyzing large 3D digital images containing branching networks. We have devised a novel segmentation algorithm (SymRG: Symmetric Region Growing) and an efficient skeleton representation method, and an automatic analysis procedure for large 3D branching networks. In this thesis we also construct visualization tools to facilitate interactions with the extracted branching networks. Finally computer-created phantom images and real 3D medical images are used to validate the proposed analysis and visualization methods. |
