Cone -beam helical CT virtual endoscopy: Reconstruction, segmentation and automatic navigation

Bruno Motta de Carvalho, University of Pennsylvania


Virtual Endoscopy (VE) is a technique in which three-dimensional (3D) data, acquired by an imaging technique such as Computerized Tomography (CT) or Magnetic Resonance Imaging (MRI), is segmented and presented in an animation so as to mimic an endoscopic examination, i.e., as if a camera were introduced into an anatomical structure. It has been shown that the detection rate of small abnormalities in VE is still below acceptable rates, pointing to the fact that there is still a lot of room for improvement in these procedures. This dissertation proposes alternative techniques for the three phases of creating a VE: image reconstruction, segmentation and animation. We propose the use of Algebraic Reconstruction Techniques (ART), and a block-iterative variation of it (block-ART), for reconstructing 3D images from helical cone-beam CT data. For efficiency and accuracy reasons, we implement ART using modified Kaiser-Bessel window functions (also known as blobs) as basis functions that are placed on the body-centered cubic (bcc) grid, instead of the traditionally used voxels of the simple cubic (sc) grid. The accuracy of the reconstructions produced by these algorithms are compared with the ones produced by a fully-3D filtered backprojection algorithm for several data sets, where is shown that ART produced more accurate reconstructions both in the absence and in the presence of realistic simulated noise. For the segmentation step, we propose a general multi-object fuzzy segmentation algorithm that segments a set by producing a map that encodes the grades of membership for all elements of this set for all objects. We report on the accuracy and robustness of our algorithm and present segmentations performed on mathematically-defined images as well as images acquired by various modalities, such as CT and MRI. As a result of our segmentation algorithm, strongest “paths” connecting a specific element of the set to any other element of the set are produced. These paths are then smoothed to produce the final navigation paths for the virtual camera of the VE animations. In order to show the applicability of our techniques we perform the three phases of generating a VE on a single data set.

Subject Area

Computer science|Biomedical research|Radiology

Recommended Citation

Carvalho, Bruno Motta de, "Cone -beam helical CT virtual endoscopy: Reconstruction, segmentation and automatic navigation" (2003). Dissertations available from ProQuest. AAI3109160.