Abstract Computed tomography (CT) images reflect the attenuation
density (gray scale) of X-rays inside the body, yet lack information about
tissue textures. For visualization purposes, the tissue textures with colors
are usually desirable. The goal of this work is to create color textures
from ground truth and synthesize them into gray scale CT images. Three-dimensional
(3D) color texture data is primordial for volume rendering to create high
quality images for visualization. Unified texture segmentation, analysis,
modeling, matching and synthesis are thus proposed toward the goal of texture
mapping from color textures of the Visible Human dataset to the CT density
dataset. First, the approach utilizes 3D region growing to segment out
the classes, each with a similar texture from the Visible Human dataset.
Then, it applies 3D wavelet transform to extract texture features from
the classes. Modeling the extracted textures is based on a multi-scale
statistical theory. Using cross-entropy criterion the modeled texture is
matched to the segmented classes of the CT dataset. Finally, the 3D color
textures of the Visible Human dataset are synthesized into the 3D CT dataset
by multiresolution sampling. This technique has been applied to our virtual
colonoscopy system in which a reconstructed colon model from a CT scan
is mapped with a generic Visible Human texture for navigation.
Acknowledgements The visible female dataset of our experiments is provided by the Visible
Human project of the National Library of Medicine. We would like to thank
Su Yang and David Zhang for their contribution in this work and Kathleen
McConnell and Lin-Huy Lu for writing help. This work is supported, in part,
by NIH Grant#CA79180 and Grant#HL51466
